OpenAI Dots Guide: 72 Use Cases and Prompts You Can Steal

OpenAI Dots gives me a way to hand an ongoing responsibility to an AI agent, then return to the same conversation as the work develops.

I’m using my Dot, Jamie, for Promptslove’s digital PR and link outreach, and the useful part is the research, follow-through, and record of what happened.

Jamie’s (My ChatGPT Dot) early September 30 checkpoint showed 13 reported outreach or submission actions and zero live links; later that evening, I checked its first live directory backlink.

In this guide, I’ll show you the setup, the practical limits, my outreach experience, 14 detailed operating prompts, and 72 more use cases with starter prompts you can adapt.

Key Takeaways

What I checked for this review

This guide reflects the information I checked on September 30, 2026. I used OpenAI’s launch material and product guides, inspected Jamie’s outreach conversation and existing task settings, and reviewed public examples from X and reader questions on Reddit.

I keep three kinds of evidence separate. OpenAI’s documentation supports product facts. Jamie’s conversation supports what the agent reported about my campaign. Public posts support what their authors said about their own experience.

That difference matters when you read a review of an agent. A confident message, a completed task, a form receipt, and an achieved business result can describe four different things.

My first outreach screenshot was captured at 19:06:56 IST on September 30. It preserves the early 0/3 checkpoint. After a later update, I checked the public listing and its destination between 20:14 and 20:16 IST, and captured Jamie’s first-link report at 20:16:30 IST. The case below records both stages of the continuing campaign.

What OpenAI Dots actually is

OpenAI announced Dots in its September 29 DevDay 2026 documentation. It describes an agent that keeps working between conversations, uses connected apps and its own computer, and asks for your judgment when needed.

The official guide calls it:

“Your dot is an always-on agent” OpenAI’s Dots guide

For me, the useful question is what responsibility I want the agent to retain. “Write this pitch” asks for a draft. “Manage this defined outreach campaign, keep the tracker accurate, and tell me when a reply or placement needs attention” describes a continuing job.

I think of the Dot as the person-shaped interface to that job. That is my explanation of the workflow, not a claim that the software is a human employee or that it makes every decision correctly.

Dots, the model, and delegated work

The underlying model and the product are different things. GPT-6 Astra powers Dots, while the Dot is the agent experience through which you assign work and review progress.

OpenAI documents delegation to ChatGPT Work and Codex. This gives you another useful distinction: the conversation with your Dot can coordinate a job while a separate task performs part of it.

Here is how I would decide where to begin:

This is a practical decision table, not a benchmark comparing products. A longer-running job still needs a clear definition of success.

“Always-on” still needs a specific job

I would not treat “always-on” as a promise of unlimited execution, immediate responses, or guaranteed success. The useful benefit is that I can describe a job that outlasts one exchange.

An editor might take days to consider a pitch. A publisher might acknowledge a submission and never add the resource. My Dot can help maintain the process, but the publisher controls the placement.

That is why my instructions need to cover what happens while I wait. The agent should record the status, watch the agreed evidence sources, and avoid repeating an action simply because an outcome has not arrived.

Who can use Dots, and what does it cost?

OpenAI documents a gradual rollout. Being eligible does not prove that the feature has appeared in every eligible account.

The launch access rules I checked are:

These details come from OpenAI’s Dots access guide. The Help Center says access may take several days.

For readers in India, the personal-plan exclusion list does not name India. Your plan, age, and rollout status still need to qualify.

The first Dot is included, but work has limits

OpenAI says the first Dot is included at no extra cost for eligible Pro and Business Premium users. Its current launch wording describes a deeper-work allowance with extended limits for the first month.

The accounting distinction is important: ordinary Dot conversations do not count toward ChatGPT usage limits, while tasks started or managed in Codex or ChatGPT Work count as usual.

I would check the actual usage view before building a heavy recurring workflow. OpenAI’s pricing guide explains shared usage and task-dependent consumption.

I did not establish a public numeric Dot-specific deeper-work quota from the checked sources. I also would not use an API token price to invent a consumer price for this feature.

My own planning question is simple: how often must this job run to make a useful decision? If a publisher replies once a week, a frequent monitor might spend work without producing a better answer.

How I would set up a Dot properly

I would start with one responsibility and one small deliverable. A manageable first job helps me see whether the agent understood my business before I add more access or recurring work.

Step 1: Create the Dot on desktop

OpenAI’s setup guide says to create your Dot in the desktop app or a desktop browser, follow the introduction, and optionally connect apps and your computer. You can add connections later.

The name and appearance are editable. I would name the Dot after its role or choose a name I can use naturally in conversation.

For a first responsibility, I would include:

Step 2: Connect the correct account

The documented app path is Settings, Plugins, the selected app, and its connection flow. Review which provider account you are authorizing.

An installed plugin and a ready account connection are different states. A connection to one account also does not grant information held only by another account.

This became a real issue in my outreach setup. Jamie initially reported a different connected mailbox from the sender I had requested.

I would confirm the sender’s identity before the first outgoing message. The test is whether the intended account is available for the intended action, not whether an email-related tool appears somewhere in the interface.

Step 3: Separate apps, messaging, and computer access

The getting-started guide treats app connections, contact methods, and your own computer as separate connections.

I would choose each one because the job needs it. If the job can use a supplied document, I can begin with that document rather than opening access to an entire working environment.

A Slack contact method gives me a way to communicate with my Dot. It is not, by itself, permission to read every business account or operate my computer.

Step 4: Understand the cloud computer

The computer guide documents a cloud computer with its own files and website sessions. Signing into a site in my personal browser does not sign the Dot into that site.

The documented inspection path is the Dot profile and Computers. Take over gives you control when needed; Return control lets the Dot resume.

If the agent says it cannot see a file or account, I would check which environment contains it. A file on my laptop and a file in the cloud computer are not interchangeable inputs.

Step 5: Connect my own computer only when needed

OpenAI’s local-computer instructions say to allow access from the Dot’s profile on that computer. One personal computer can be connected; it needs to stay online with ChatGPT open for local work.

Offline and Revoke access mean different things. The first describes availability; the second removes the grant.

For my outreach example, the profile showed Jamie’s computer. I am not treating that screenshot as proof that my own Mac had been connected.

Step 6: Review the profile and the first output

In Jamie’s profile, I could see the computer, recent activity, and outputs. It also showed the hourly link-building task and a progress report scheduled for October 1 at 9 AM.

I would inspect the output itself. A table should contain useful, specific evidence. A pitch should accurately represent the product. A promised attachment should open and contain the expected work.

The best first result is small enough that I can judge it. Five well-explained prospects teach me more about the agent’s understanding than a huge list with vague descriptions.

How I brief an ongoing responsibility

A useful brief gives the agent a reason to choose its next action. I want the Dot to know what matters, what information it should use, and what counts as completion.

I would put these seven items into an ongoing brief:

For outreach, “get me links” leaves important decisions unstated. Which pages deserve promotion? What counts as a relevant publisher? What happens if a form has no receipt? Can the agent spend money or accept new terms?

I prefer a brief that resolves those decisions before the agent reaches them. That does not remove built-in safeguards, but it reduces ambiguity within the work I actually want done.

Define completion before starting

I would make the completion check observable. “Produce five pitches with exact recipient routes and factual personalization” is reviewable.

“Grow the business” is a larger objective that needs smaller milestones. An agent can work on the milestones while I still judge whether they contribute to the business.

I also want an exception path. If a needed source is missing, the Dot should say what is missing and continue useful work that does not depend on it.

Tell the Dot when to interrupt me

I do not need an update for every page it reads. I need an update when something changes my next decision.

For a campaign, that might be a reply, a verified placement, a missing account, an uncertain send, or a deadline that cannot be met with the current evidence.

This makes the interaction easier to manage. I can inspect the tracker for routine activity and use notifications for the moments that need my attention.

How I’m using Jamie for outreach

My current use case is digital PR and relevant link acquisition for Promptslove. I asked Jamie to understand the business, identify suitable opportunities, prepare pitches, and keep working on the campaign.

I am focusing this review on that outreach work. I am not using an abandoned earlier plan as evidence that Jamie manages every part of my marketing.

The first job was choosing a useful destination

Jamie proposed promoting resources that fit the publisher’s audience. Its plan included useful tools, product pages, research, tutorials, and templates.

The specific destination matters. An editor collecting AI video resources might care about a relevant product or tutorial. A general homepage can make the editor do the work of finding the useful resource.

Jamie’s proposed first deliverable was 30 qualified prospects, five priority landing pages, and five pitch drafts. Those numbers describe its initial plan; they are not proof that every planned item was independently completed or accepted.

For me, a qualified prospect needs an explanation I can inspect. I want the exact page, the audience fit, the resource to suggest, and an official way to contact the publisher.

Jamie checked the research before pitching it

One possible PR angle involved my X-post study. Jamie asked for the dataset and analysis scripts behind the 21,864-post report, rather than treating a large number as sufficient evidence.

It then reported an inconsistency: page 31 described a corrected-link matched analysis, while page 129 said that analysis had not been rerun.

I am describing Jamie’s finding in the conversation. This review does not independently reproduce the dataset analysis or establish which statistical claim is correct.

The useful behavior was the decision to keep disputed statistics out of the pitches. Jamie later reported proceeding with commentary pitches while the research evidence remained unresolved.

This is a practical lesson for anyone using a Dot for PR. A weak research claim can affect more than one draft if the agent saves it as business context and repeats it across a campaign.

I would give the agent a claims ledger. Each reusable claim should have a source, scope, date, and status such as approved, uncertain, or excluded.

A connected mailbox was not automatically the intended sender

I specified a business sender for the campaign. Jamie initially reported that a different mailbox was connected and that it could not verify the requested sending address.

The conversation then shows the correct account being added and Jamie reporting that it was ready to use that sender. It also said it would check existing conversations before sending.

I would keep that identity check in every outreach workflow. A technically available mailbox can still be the wrong identity for a brand, product, or recipient.

The public prompts below use [approved business sender] in place of the real address. You should replace it with the specific account you intend to use, then verify that account in your own setup.

I authorized action, but the campaign still needed boundaries

The conversation records my authorization to proceed with the outreach. That does not mean every possible business action was defined in advance.

For example, a free listing can introduce a terms step. A contact form can introduce a CAPTCHA. A new prospect can require a different pitch or a claim that was not reviewed in the first batch.

My practical preference is to define the authorized recipients, sender, resource, message scope, and limits early. Clear instructions help the Dot continue the work I actually approved.

OpenAI explains that custom permissions and instructions do not override built-in safeguards. I would not write a prompt that promises to remove every review or handoff.

What Jamie reported doing

The following table describes the routes in Jamie’s outreach history. It is an inventory of agent-reported actions and statuses, not an independent audit of every recipient’s inbox or publisher’s site.

The row count does not equal the confirmed-action count. This inventory includes a blocked route and an uncertain attempt, and it spans several updates.

The historical activity checkpoint is 13 reported confirmed outreach or submission actions, with zero of three verified live links at 19:06:56 IST on September 30.

Jamie later reported its first live backlink and progress of 1/3, which I checked on the public page. I have not inferred a new total of confirmed outreach actions from that later update.

That is useful progress to inspect, but it is not a result I would turn into a claim about rankings, sales, or campaign ROI.

The uncertain Geekout submission was a useful test

Jamie reported that it clicked Send once and received no confirmation. It also reported that it had not retried, to avoid creating a duplicate.

I like the distinction between “I clicked the button” and “the site acknowledged the submission.” A form that clears can be ambiguous, especially if the success notice disappears before a screenshot.

My next step would be evidence reconciliation. Check an authorized sent record, a confirmation email, or the publisher’s submission status if one exists.

If the outcome remains unknown, keep it unknown in the tracker. An accurate uncertain status is better for the campaign than a confident label that encourages a second send.

A blocked route should not freeze the whole job

Jamie reported a verification block at Future Tools and continued with other opportunities. The conversation linked OpenAI’s cloud-browser guidance on blocked sites.

That guide describes a connected-computer task as a possible alternative for some blocked cloud work. It does not guarantee that every site will become usable.

I would ask for the exact blocker and a supported alternative. If the route needs my action, the Dot can prepare the pitch and field values while it continues research elsewhere.

I do not want a stalled form to become an excuse for repeating the same failed interaction. I also do not want a blocked route counted as an outreach result.

Pending review is a real status

Jamie reported that the SaaSHub listing was pending review and that the site said review could take up to 32 days. That is a report about this submission flow, not a general turnaround guarantee.

A campaign deadline does not change the publisher’s timetable. If a listing is pending, I would record the expected next check rather than treating it as a link already earned.

This also changes how I choose opportunities. A three-day target needs some routes that can plausibly publish within that window, while slower routes may still be worth maintaining for later.

Update: I checked our first successful backlink

Later that evening, Jamie reported its first live placement at Full Stack Journalism and changed the campaign progress to 1/3. That gave me a concrete result to inspect.

I checked the public tool directory, opened the matching record, inspected its website link, and loaded the destination. Here is what I documented:

The destination page loaded without a login handoff during this check. That establishes that I could view the intended landing page. It does not test every feature of the tool.

For my tracker, this is a live contributed directory backlink, with editorial review still unresolved. I would record that type explicitly so I can compare contributed listings, reclaimed mentions, and earned editorial citations later.

The nofollow detail matters. Google explains that qualified links are generally not followed and that a page can still be discovered through other routes. I would not turn this placement into a promise of ranking credit, traffic, or AI citations. Those results need their own measurements.

I can independently confirm the live presence, link attribute, and destination at the time I checked. The account of how the submission was made comes from Jamie’s conversation. I have not independently audited the publisher’s internal review or the entire submission history.

How you can inspect this example yourself

This is the part of the campaign I wanted to see: a public result I can open and inspect. I will keep the earlier 13-action checkpoint because it shows the work before the first placement, rather than rewriting the campaign history around the latest result.

Jamie also reported sending RanknestAI to AI-Tools.ma’s curator for review after this update. I am keeping that as a reported submission, with no live placement confirmed in this review. The campaign was continuing toward the remaining two links.

How I measure outreach without fooling myself

I want the tracker to separate preparation, action, response, and publication. That makes it easier to tell whether the campaign needs better targeting, a clearer pitch, more patience, or a different resource.

Here is the status ladder I would use:

I would keep each status beside its evidence reference. If I cannot open or identify the evidence, the status needs qualification.

The tracker fields I would keep

A useful tracker should answer what happened, why the target fits, and what comes next. It should also help the agent avoid repeating work.

I would not fill the table with invented authority scores or traffic estimates. If those inputs are unavailable, topical fit and the visible editorial page still provide useful qualification.

A live link needs its own check

My live-link definition is a new public link on a relevant third-party page to the approved destination, verified at the time of review.

That excludes an email pitch, a form receipt, a promised listing, an owned publication, a pre-existing link, and a plain-text mention.

For the campaign target, I would also avoid counting the same placement twice after a redirect or URL variation. A clear resource and publisher record makes that easier to reconcile.

If the page is inaccessible, I would mark it unverified. I would not infer publication from a positive reply alone.

SEO and GEO need their own evidence

Google’s link-spam policy covers manipulative link practices, including automated services used to create ranking links and improperly qualified paid links.

My recommended workflow is relevant editorial research and truthful outreach. I would avoid bulk directory blasts, comment spam, link farms, and promises to purchase ranking credit.

Even a legitimate placement does not prove an improvement in search traffic or AI citations. I would track referral visits, leads, search performance, and observed AI mentions separately, using the available analytics and a clear date range.

The presence of a backlink is one observation. Its commercial effect is another question that needs its own evidence.

The hourly task I found in Jamie’s setup

Jamie’s profile showed Promptslove Link Building as an hourly task. The saved instructions targeted three legitimate, relevant, verified new external links by approximately 18:51 IST on October 3, 2026.

The instructions also covered checking replies and receipts, reconciling contact history, avoiding disputed research claims, excluding pending or owned links, and reporting meaningful changes.

A written deadline and a saved end date are different

The important setting in the screenshot is End repeat: Never. The October 3 target appears in the instructions, while the scheduler’s end control does not show that date.

I would not describe this as a task that automatically ends on October 3. If I wanted a strict scheduled stop, I would ask the Dot to set the actual recurrence end date and then inspect the saved control.

In this existing task, the instructions say to report the outcome at the deadline and continue authorized follow-through unless I stop or change the campaign. That explains why a deadline in the text should not be treated as a hard stop.

Hourly monitoring should not mean hourly chasing

An hourly check can look for a new reply or a changed publisher page. It does not need to send an editor a message every hour.

I would define a separate follow-up cadence for outgoing communication. I would also specify that declines, opt-outs, and uncertain earlier attempts should suppress automatic follow-up.

My preferred notification rule is a meaningful change: a reply, a verified placement, a decision I need to make, or an actual blocker. Routine checks can update the record without producing a new message every time.

Confirm what was actually saved

OpenAI’s recurring-task guide says to specify the job, schedule, time zone, end date, notification conditions, and result destination, then review the saved task in Scheduled.

For my own review, I would ask the Dot to read back the task’s exact settings. That gives me a checkable answer:

The answers should agree with the saved settings. If they do not, I would resolve the difference before relying on the recurrence.

Permissions, memory, and the controls I would check

The useful question is whether the agent can perform the particular job with the particular account and authority I intended. A general statement such as “you have permission” can leave the recipients, account, budget, or action unclear.

Connection, permission, and instruction are different

I would check these layers separately:

OpenAI’s app-permission guide describes read and action settings that can vary by account. These controls do not connect accounts or override provider and workspace restrictions.

I would put an exact sender and target list in an outreach authorization. That makes the permitted batch easier to continue and easier to audit.

Use custom rules for a durable action boundary

The documented path is Settings, Personalization, Permissions, Custom rules. The guide lists choices for acting without asking, acting when you say so, asking before acting, and handing the action to you.

I would use a durable rule for an action boundary that matters across jobs. My writing voice, preferred report format, and campaign details can stay in the working brief.

For example, I might want a consistent review boundary for messages to new external recipients. The campaign instruction can then identify the specific approved batch.

I would avoid a pile of vague overlapping rules. “Manage everything,” “always ask,” and “never interrupt” do not tell an agent what to do when the same action matches all three.

Treat memory as a working record

OpenAI distinguishes conversation context, relevant ChatGPT memory, and a Dot’s persistent notes. The notes are not a complete transcript.

For my campaign, the useful memory is operational: approved product facts, tone preferences, exclusions, unresolved evidence, and the current tracker.

I would periodically ask the Dot to summarize that record. If a product description or campaign target changes, I want the active brief to change with it.

Changing a ChatGPT memory setting does not necessarily change the Dot’s existing notes.

An old statement can become a new problem if the agent continues to reuse it. The study inconsistency in my outreach case is a good reason to mark uncertain claims explicitly.

Check data controls without assuming everything stays local

OpenAI says data settings apply to eligible Dot conversations and tasks. It does not train directly on proactive research and private notes, but applicable settings govern information that enters an eligible conversation or task.

I would not describe this as “everything stays on my computer” or “nothing is retained.” Those claims are stronger than the documented setup supports.

My practical approach is to give the job the relevant context and keep unrelated personal material out of the brief. The prompts in this guide use placeholders rather than my actual business sender.

Disconnecting an app also does not erase information the Dot already obtained. I would check the relevant deletion controls if removal is the goal.

Stop the parent, the task, and the recurrence deliberately

The controls guide separates pausing the Dot, stopping delegated work, and canceling recurring work. Ending a voice call also does not stop assigned work.

I would ask the Dot to identify which active job and saved schedule it is stopping. I would then inspect the relevant task and recurrence.

An already delivered pitch remains delivered. Stopping future work cannot recall the email or undo the publisher’s decision.

The documentation uses Delete on the web guide, while the Help Center also uses Reset. Follow the actual confirmation shown in your account.

Messaging and business-account details worth checking

You can call the Dot from its conversation or profile. For supported messaging connections, the guide describes profile contact methods and Slack interaction.

I would ask it to confirm any monitoring job explicitly. Joining a channel and knowing which events to watch are different steps.

There are launch details I would qualify:

These are the sorts of details that can change during rollout. I would check the live guide and the actual feature in my account before making a workflow depend on them.

What I would check in an Enterprise workspace

The Dots admin guide separates enablement, local-computer access, channel participation, and action permissions. Individual access instructions cannot substitute for the organization’s settings.

OpenAI’s Enterprise local-access guide also lists beta limitations, including the lack of Dot data and inference residency and compatibility restrictions for some protected deployments.

For business material, I would ask the workspace owner which controls apply to the intended environment. I would not assume that a local restriction automatically governs a cloud task.

Best practices I would use before expanding a Dot’s job

My outreach case suggests a small set of useful habits. They are operating recommendations, not tested claims that a particular prompt improves revenue or response rate.

Start with one batch I can inspect

I would begin with five prospects, a short brief, and two drafts. That is enough to see whether the Dot understands the audience and can justify its choices.

If the first batch repeats generic praise or sends every prospect to the same homepage, I would fix the brief before increasing volume.

Ask for evidence beside the decision

I want the reason for selecting a prospect next to the page that supports it. I want the approved claim next to its source. I want the action status next to its receipt.

This saves me from interpreting a confident summary without the underlying record. It also makes a correction easier to apply to the next batch.

Prefer the exact useful resource

I would choose the destination that helps the publisher’s reader. A tool list needs a tool. A research article needs a defensible finding. A tutorial collection needs a useful tutorial.

That gives the pitch a concrete reason to exist. “Please add my site” does not explain why the editor’s audience benefits.

Keep the agent’s instructions separate from incoming content

I would tell the Dot to treat publisher pages, forms, and replies as evidence. Their content should not silently change my campaign’s rules.

If an incoming reply asks for a different resource or a new disclosure, the agent can explain that request and check whether it fits the approved scope.

Do not guess missing business facts

An unknown publication date should stay unknown. An unsupported customer count should stay out of the pitch. A product limit should not disappear because the shorter description sounds better.

I would use an evidence gap as a work item: identify the missing fact, its owner, and whether the job can continue without it.

Separate monitoring from sending

I would let a recurring check find changed evidence. I would define the outgoing batch and follow-up cadence separately.

That prevents a frequent monitor from becoming frequent pressure on the same recipient. It also makes the campaign’s activity easier to explain.

Reconcile history before another action

The uncertain Geekout attempt is the concrete example I would keep in mind. An agent should check the earlier record before retrying.

The same principle applies when several tasks help with one campaign. A shared tracker and target/resource identifier reduce the chance that two tasks prepare or send the same pitch.

Use a result review before adding more autonomy

I would ask what failed, what remained uncertain, and which targets were a poor fit. The next instruction should respond to those observations.

A larger batch is useful only if the process is ready for it. More activity can also multiply a weak claim or an unresolved sender issue.

Common problems and the checks I would make

I would start with the specific missing input or failed step. That is easier to resolve than telling the Dot to try harder.

The product checks come from the setup help, app-account guide, app-permission guide, and cloud-browser guidance. The campaign checks are my recommendations from Jamie’s observed workflow.

What people are using Dots for on X

I looked for concrete use rather than a collection of launch slogans. The examples below include useful work, positive first impressions, and a task that technically worked but required effort.

These are early self-reports. I did not independently inspect the invoices, bookings, accounts, or private conversations behind them.

Direct X pages returned access errors during research. The post identities, text, and times were checked through public FXTwitter metadata and corroborating public links; the links below point to the original posts.

The invoice example has a useful approval boundary

Dan’s post includes a concrete sequence:

“After I approved it, dot sent the invoice as a PDF.” Dan McAteer’s post

I would use the same preparation-and-review idea for many business tasks. The agent can gather the material and prepare the action while I inspect the parts that matter.

That invoice does not prove that Dots improves outreach conversion. It shows one person’s reported workflow.

The dessert example is a useful reality check

Jane wrote:

“The cannoli arrived. It technically worked.” Jane Manchun Wong’s post

If a job takes more of my attention than doing it myself, completion alone does not make it a good delegation candidate. I would test the full cost in attention, checking, and correction.

One dessert order is not a general speed benchmark. It is a useful reminder to judge the experience as well as the final artifact.

Early enthusiasm can coexist with an unfinished experience

Matt’s qualification is clear:

“But the UX needs a lot of work before I can make it my daily driver.” Matt Shumer’s post

My interpretation is that asynchronous preparation can be a good place to begin. I can review a finished brief or draft without depending on the agent to make every interaction feel instant.

I would not treat this selected set as a survey of user satisfaction. Early-access testers are overrepresented, and an affiliated example deserves its affiliation label.

My review: what looks useful and what I cannot claim yet

The most useful part of Jamie’s observed outreach is the continuity of the job. The conversation records research, a campaign pack, a sender issue, several submission routes, uncertain outcomes, and a continuing status update.

I can see why that is more useful to my campaign than an isolated pitch draft. It gives me a process to inspect.

I have not measured the time saved, delivery rate, reply rate, or incremental revenue. I also have not independently audited every reported action.

For me, the next review should preserve the first placement’s evidence, reconcile the tracker, inspect replies, and decide whether the next batch needs a different angle. One documented link gives me an outcome to inspect while the campaign continues.

14 detailed prompts I would reuse for outreach

These prompts were written for this guide. They are proposed operating instructions, not a claim that I pasted all of them into Jamie.

Replace the bracketed fields with your actual business inputs. Start with preparation, inspect the output, and authorize a clear batch when you are ready for an external action.

I would use them in order when building a new campaign, then return to the relevant prompt when a particular problem appears.

Prompt 1: Set the outreach role and boundaries

Prompt 2: Build a claims ledger before a PR pitch

Prompt 3: Research a focused prospect list

Prompt 4: Reclaim an existing mention

Prompt 5: Verify the sender and contact route

Prompt 6: Write a pitch in my voice

Prompt 7: Prepare a form submission and evidence plan

Prompt 8: Authorize a specific small batch

Prompt 9: Handle a send with an uncertain outcome

Prompt 10: Deal with a blocked site

Prompt 11: Review replies and prepare a follow-up

Prompt 12: Verify a live placement

Prompt 13: Maintain a useful progress summary

Prompt 14: Review the workflow before expanding it

72 Dots use cases with starter prompts

These are workflows I would try with a dot, followed by prompts you can adapt. They are proposed uses based on documented research, document preparation, connected-tool work, computer use, delegation and recurring tasks. I have not tested all 72, and this list does not describe work Jamie has already completed.

Connected work requires access to the relevant sources and permission to use them.

A named app, folder or website in a prompt is an input you provide, rather than a promise that a particular integration is available. If access is missing, ask the dot to explain the gap and work from files you provide.

Replace the bracketed inputs before pasting a prompt. Start with one small batch and inspect the result. For recurring work, include your time zone and an end date, ask the dot to confirm the saved instruction, and inspect Scheduled.

I wrote these examples to keep messages, purchases and changes to live systems for your review.

Marketing and content

1. Turn one interview into a content pack

I would give my dot one transcript so I can review a coordinated set of content instead of starting each draft separately.

2. Build a useful interview brief

I would use a dot to find the questions my audience needs answered before I record an interview.

3. Draft a focused newsletter issue

I would ask for one issue built around a clear reader problem, with evidence I can check.

4. Create a working voice guide

I would turn my own edits into a short guide that helps the dot produce more consistent drafts.

5. Plan a campaign around real capacity

I would use a dot to connect campaign goals with the time, people and assets I actually have.

6. Prepare a customer email sequence

I would ask for an email sequence that reflects the product's real behavior and gives me clear review points.

SEO, AI search and digital PR

7. Find gaps in a defined content set

I would compare a small set of my pages with the questions I want to answer for readers.

8. Refresh an article's facts

I would have the dot check an aging article before I update it in my publishing system.

9. Review page basics from supplied material

I would use a dot to catch obvious page problems without pretending that a quick review is a full technical audit.

10. Propose useful internal links

I would ask the dot to find links that help a reader take the next sensible step.

11. Check coverage of buyer questions

I would review whether my content answers real questions clearly before discussing visibility in AI search.

12. Prepare a digital PR evidence pack

I would ask a dot to organize my evidence before I approach a journalist or publisher.

Sales research and customer-support preparation

13. Research a small prospect shortlist

I would start outreach with a few qualified companies and evidence I can inspect.

14. Draft evidence-based prospect messages

I would use a dot to prepare messages that connect a real prospect need with an offer I can substantiate.

15. Create a meeting-ready account brief

I would let the dot summarize the account history so I can focus on the conversation.

16. Triage replies for my decision

I would ask my dot to organize a small reply batch and draft responses I can approve.

17. Build a proposal evidence matrix

I would use a dot to find where my proposal needs proof before I make a commitment.

18. Prepare support answers from approved knowledge

I would have the dot turn recurring support questions into draft answers for my team to review.

Business and personal administration

19. Prepare a missing invoice

I would ask the dot to assemble an invoice draft from records I already trust.

20. Organize one batch of receipts

I would use a dot to reduce receipt sorting before I review the numbers myself.

21. Review recurring subscriptions

I would let the dot show me recurring charges and the evidence behind possible cancellations.

22. Draft overdue-invoice reminders

I would prepare a clear review queue before following up on unpaid invoices.

23. Prepare a vendor renewal decision

I would compare the renewal notice with the work I need the vendor to support.

24. Sort a weekly admin backlog

I would ask my dot to turn a messy admin list into a realistic set of next actions.

Meetings and project coordination

25. Prepare an agenda from the actual work

I would build a meeting around decisions rather than a generic status list.

26. Extract decisions and action items

I would let a dot turn a meeting transcript into a record I can correct and share.

27. Map launch dependencies

I would use a dot to find the dependencies that could make my launch plan slip.

28. Prepare a weekly project update

I would ask for a short update that helps people understand progress and the decisions I need to make.

29. Assess a proposed scope change

I would ask the dot to make the effect of a scope change concrete before I approve it.

30. Build a stakeholder review pack

I would give reviewers the material they need to make a specific decision.

Research and learning

31. Create a source-checked research brief

I would use a dot to collect evidence around one question before I form an opinion.

32. Critique a research paper

I would ask a dot to help me understand a paper's claim, method and limits.

33. Watch a small research topic

I would ask the dot to alert me when a new source changes what I need to know.

34. Build a study plan around my time

I would use a dot to turn a learning goal into sessions I can complete.

35. Practice from material I provide

I would ask the dot to help me test my understanding and focus on my weak spots.

36. Compare options against explicit needs

I would research a purchase decision using criteria that matter to me rather than a generic ranking.

Coding, product design and quality checks

37. Investigate one reproducible bug

I would give the dot a specific bug and ask it to show me what actually fails.

38. Prepare a small tested patch

I would ask my dot to complete a bounded code change that I can review before merging.

39. Check a screen for accessibility issues

I would review one interface before I repeat the same pattern across my product.

40. Prototype one product flow

I would use a dot to make a small idea concrete enough for people to react to.

41. Check a defined regression path

I would ask my dot to verify the actions a change could affect, using a clear test environment.

42. Align documentation with a shipped feature

I would have a dot identify documentation that no longer matches the product.

Spreadsheets and data reporting

43. Inspect a dataset before analysis

I would check the data first so I do not build a convincing report around bad inputs.

44. Explain a small set of KPIs

I would ask the dot to connect the numbers with the decisions I need to make.

45. Review budget variance

I would use a dot to show me where actual spending differs from the plan.

46. Build a simple scenario model

I would ask the dot to show how a decision changes when my assumptions change.

47. Compare customer cohorts

I would use a dot to find differences between groups while keeping the definitions clear.

48. Summarize a survey with traceable evidence

I would ask the dot to show both the numbers and the comments behind the main themes.

Documents and knowledge organization

49. Build an index I can actually use

I would ask a dot to show what is in a folder before I reorganize it.

50. Draft a procedure from real examples

I would turn a repeated task into instructions that another person can follow.

51. Prepare an internal FAQ from policy documents

I would make existing rules easier to find without giving the dot authority to create new ones.

52. Create an onboarding reading path

I would help a new teammate understand the job without dropping a pile of documents on them.

53. Plan a document migration

I would use a dot to plan where files belong before making changes that affect other people.

54. Preserve the reasoning behind a decision

I would ask my dot to record why I chose an option so I can revisit it later.

Travel and event planning

55. Draft a trip itinerary with current sources

I would plan a trip around my actual dates, budget and preferences, then make the bookings myself.

56. Compare venues for one event

I would ask a dot to narrow the venue search before I make enquiries.

57. Build an event running order

I would use a dot to turn the event plan into a schedule my team can review.

58. Prepare a speaker briefing pack

I would make the event's expectations clear before I send anything to a speaker.

59. Check the paperwork for an upcoming trip

I would ask a dot to organize the documents I already have and flag missing information.

60. Plan event access and guest needs

I would check practical arrangements before guests arrive and discover a problem I could have caught.

Household and everyday planning

61. Build a grocery list from what I have

I would start meal planning with my pantry instead of asking for another expensive shopping list.

62. Organize routine home maintenance

I would give my dot the manuals and dates so I can keep track of routine upkeep.

63. Find a thoughtful gift shortlist

I would use a dot to compare a few ideas against what I know about the person.

64. Plan a manageable house move

I would turn a move into a small set of weekly actions I can complete.

65. Draft a fair household task plan

I would use a dot to make routine work visible before discussing how we divide it.

66. Prepare my weekly personal review

I would ask my dot to help me see what needs attention without turning every detail into a notification.

Recurring monitors with clear notification rules

67. Watch a few competitor pricing pages

I would ask for changes I can verify, with alerts only when something relevant moves.

68. Monitor important website pages

I would keep an eye on a small set of pages so a visible error does not sit unnoticed.

69. Watch defined project blockers

I would let a dot monitor a few known dependencies while I concentrate on other work.

70. Flag a repeated support problem

I would ask my dot to notice when the same problem starts appearing more often in the data I provide.

71. Watch dates that need my decision

I would ask for a reminder when a deadline needs action, rather than another daily recap.

72. Check my outreach review queue

I would use my dot to keep a small outreach queue ready for review while I control the sending decision.

How I would choose the first one

I would choose a job with clear inputs, a visible output and a quick way to judge the result. A five-company research table, one invoice draft or one transcript content pack gives you a better first test than an open-ended instruction to run your business.

You can expand the responsibility after reviewing the first batch. Add specific approval rules only when you understand the actions, recipients and limits involved.

Connections, permissions, and built-in safeguards still apply.

Frequently Asked Questions (FAQs)

These questions reflect real reader discussions about what Dots means and regional access, non-coding uses, plan eligibility, usage accounting, and day-one speed and friction.

What is OpenAI Dots?

Dots is an ongoing agent in ChatGPT powered by GPT-6 Astra. You assign a responsibility and review the work as it develops.

What can I use Dots for besides coding?

Public examples include invoice preparation and event organization. The 72 suggested workflows above also cover research, content, admin, documents, reporting, and planning.

Which ChatGPT plans include Dots?

The checked launch guidance names eligible Pro tiers, Business Premium, and administrator-enabled Enterprise access. Check your account because rollout is gradual.

Is Dots available in Europe and the UK?

The personal Pro access rules exclude the EEA, UK, and Switzerland. Business access has separate rules, so check the relevant plan rather than applying the personal restriction to every account.

Does Dots count against my ChatGPT usage?

Dot conversations and the Codex or Work tasks it manages have different accounting. Check the deeper-work allowance and downstream task usage in your account.

Is Dots fast enough to use for everything?

There is no basis in this review for that claim. Early users report both useful administrative work and friction, so I would test one specific job and judge the result, attention required, and corrections.

Final Thoughts

For me, OpenAI Dots is most useful when I can name the responsibility, provide reliable inputs, and inspect a real output. Jamie’s campaign now has one checked live directory backlink, with its nofollow attribute and editorial-review status recorded. I would start with one of the prompts above, review a small first batch, and expand only after the evidence supports that decision. Give your Dot a clear job and a clear completion check, then judge what it actually delivers.