Learning how to use ChatGPT Dots starts with choosing useful work, not connecting every app you own. The first assignment should be small enough to review properly and valuable enough to tell you whether the approach helps. This guide takes you from checking access to preparing the brief, reviewing a first result and improving the process. It uses a hypothetical website content task so the advice stays concrete without suggesting that a particular WordPress integration is built into Dots.
The setup steps described here reflect OpenAI’s guidance checked on 1 October 2026. Interface labels and access conditions can change. If the feature is not present in your account, begin with the preparation sections and check OpenAI’s current availability information. Do not treat eligibility as confirmation that your specific account is already enabled. The objective is to create a first task that can be assessed fairly, whether you can run the pilot today or need to wait for access.
Step 1: choose a result you can inspect
A good first result is a draft that helps a person make a decision. For a service business, that might be a proposed FAQ brief using questions the owner has already approved for sharing. For a retailer, it might be a comparison of three draft category descriptions. Avoid beginning with a request to manage all marketing, handle customers or optimise the whole website. Those assignments contain too many decisions for a useful first test and make it hard to identify why a result succeeds or fails.
For this example, the goal is a draft FAQ brief for a maintenance service. The business wants to know which questions need clear answers on its website. The source material consists of a service outline and a small set of anonymised enquiry questions. The reviewer is the owner. Completion means an organised list of questions, suggested answers grounded in the outline and an explicit list of facts that still need confirming. No publication or customer communication is included in the assignment.
Step 2: create your dot if access is available
OpenAI’s getting-started guide says to open Dots in ChatGPT on a desktop app or desktop browser, follow the introduction and connect relevant apps or skip that step. Desktop setup also offers the choice of connecting your computer. Personalisation can be changed after setup. Follow the official setup instructions for the account you are using. These are the product steps; the rest of this guide explains how to prepare and evaluate the work you assign.
Before connecting anything, write down which information the pilot requires. If you can provide a suitable document directly, that may be sufficient for an initial draft task. If you need a service connection, review which account it uses and the available actions. Connect the right source for the actual assignment rather than guessing that additional access will improve the result. A small, known source set makes it easier to determine whether the assistant has understood the business question you are testing.
Step 3: tidy the source material
Prepare one current service outline. It should state what is included, what is excluded and which statements have been approved. Remove draft prices that are no longer valid and label anything provisional. Give the document a meaningful name and a review date. If an older document must remain available for comparison, explain that it is historical. A task cannot be evaluated cleanly when the supplied material silently presents several different versions of the same offer as current.
Then prepare the enquiry questions. Remove personal details that are irrelevant to the FAQ brief and preserve enough wording to show what the customer was trying to understand. If several questions mean the same thing, you can still retain them as evidence of a repeated theme. Do not add an answer unless the approved service outline supports it. The purpose of the pilot is partly to reveal missing information, so an honest gap is more useful than a guessed explanation that makes the draft look complete.
Step 4: write the first task brief
Here is an original example you can adapt: “Prepare an internal FAQ brief for our maintenance service. Use the supplied approved service outline and anonymised enquiry questions. Group repeated questions, suggest plain-English answers where the outline supports them, and list anything requiring confirmation. Provide a draft for my review. Do not publish, contact customers, change prices or invent service commitments. Begin by summarising the scope and identifying missing information.” Replace the service and sources with the actual pilot you want to run.
This brief defines the outcome, evidence and boundary. It also tells the assistant how to begin. Asking for a scope summary is useful because misunderstandings can be corrected before a long draft is produced. A brief does not need complicated terminology. Clear instructions such as “quote the source section for each answer” often contribute more than a long description of an expert persona. The business needs traceable reasoning and usable text, not a performance of authority that is unsupported by the records.
Step 5: inspect the scope before expanding the task
Check the assistant’s understanding against the assignment. Has it identified the maintenance service correctly? Is it using only the intended material? Does it understand that the output is internal? Has it distinguished drafting from publication? If a source is unavailable, let the task report that directly. Do not quietly substitute a different document without considering whether the result will still answer the original question. A clear access failure is easier to solve than a confident draft built from the wrong information.
You might discover that the outline does not explain cancellation or geographic coverage. That is useful feedback. Ask the owner to settle those facts, then provide the approved answers as new source material. Keep the additions specific: “For this pilot, coverage is limited to the named towns in the updated outline.” Avoid turning the correction into a general instruction to assume whatever the assistant thinks customers would prefer. The purpose of the task is to describe the real service accurately.
Step 6: review the first draft in layers
Review facts before polishing language. Check whether each suggested answer matches the approved offer. Pay attention to dates, quantities, inclusions and conditions. Then assess whether the response answers the customer’s actual question. Only after those checks should you improve tone and formatting. A beautifully written answer that describes the wrong service creates more work, because the underlying claim needs correcting before the editor can make it suitable for the website.
Use three review labels: approved, needs a factual decision and needs an editorial revision. For example, an answer explaining how to request a quote may be approved. An answer about emergency response time may need an operational decision. An answer that is accurate but too long may need editing. Separating these categories stops a copywriting revision from being mistaken for a commercial approval. It also shows the next person exactly what they are expected to do with the draft.
Step 7: correct the process, not just the paragraph
If the first draft invents a service commitment, explain the error and add a reusable instruction: “Where the outline is silent, identify an unanswered question instead of supplying a likely answer.” If it produces too many overlapping questions, provide a sample of the grouping you want. If the tone is too formal, supply a short approved paragraph as a writing reference. Each correction should address an observable issue, so you can judge whether the next result has actually improved.
Avoid responding with “make it better” unless you also explain what better means. In the FAQ example, it could mean shorter answers, clearer distinctions between services or more obvious unanswered questions. Give the assistant the criterion and, where possible, a concrete example. Retain the improved brief for the next task. This makes the pilot a learning process and reduces the chance that someone has to reconstruct the same instructions when the work is handed to another reviewer.
Step 8: verify where the result was saved
A task is not complete merely because a message says that work has been done. The business should be able to open the intended result and inspect it. If the assignment asked for a file, check that the file exists and contains the final version. If it asked for changes to a draft document, open that document and verify the revisions. If access prevented saving, retain the prepared content and identify the blocker. Preparation and successful delivery are separate outcomes.
For the FAQ pilot, the final handover should identify the draft, its source outline and the questions still waiting on a person. Give the reviewer a clear next action. It might be “confirm cancellation wording before the website editor prepares the page”. Do not describe the FAQ as ready for public use if unresolved commercial facts remain. This final check helps the business avoid losing useful preparation in an ambiguous conversation that nobody can confidently turn into a finished customer-facing result.
Step 9: test with an ordinary second example
Once the first task is acceptable, repeat the method with a different but comparable source set. You might use the same service outline with a new batch of approved questions. Keep the outcome and review method consistent so you can see whether the process is improving. Do not simultaneously add a new tool, a new output format and public publishing authority. If the result changes, you will not know which change caused the difference, and the pilot will be harder to assess.
Record how much time preparation and review take. Also note whether the assistant identified genuine gaps or created extra ones. A useful process may save the owner attention even when it does not reduce every minute of writing time. Conversely, a fast first draft may be a poor investment if checking it takes longer than producing the brief directly. Compare total effort and the quality of the handover, using the same acceptance question each time.
When to introduce an ongoing responsibility
Only expand into continuing work when the discrete task has a clear purpose and a reliable review method. For the example above, an ongoing responsibility could be preparing a fresh internal FAQ brief from an approved source whenever the owner supplies new questions. This is a suggested workflow, dependent on the tools and permissions available. Define what counts as new material, how duplicates should be handled and when the owner wants a result. Ask the product to confirm the actual assignment it has saved.
If a fixed schedule matters, specify a local time zone, duration and destination, then verify the saved schedule. OpenAI’s recurring-task guidance explains the product requirements. Do not assume that describing a weekly goal creates a schedule or that connecting a service starts monitoring it. For a first pilot, a short trial period can be easier to review than an indefinite responsibility that nobody remembers to reassess.
Troubleshooting a first task
If the feature is absent, check the account’s access conditions before repeatedly changing settings. If the task cannot read a source, confirm the intended document and connection rather than giving broader access automatically. If the output is unrelated, narrow the brief and supply a representative example. If the draft is plausible but inaccurate, ask for evidence beside the claims and remove unsupported conclusions. The right fix depends on the problem, so start by identifying which stage failed.
If the work appears to stop, inspect the task’s reported status and any request for input. A missing decision is different from a missing connection. Give an answer to the actual question rather than restarting the whole assignment and creating duplicate work. If the task reports success, inspect the destination. Keep a note of unresolved problems before trying another run. This allows a reviewer to distinguish completed work, blocked work and an output that still needs correction.
Frequently asked setup questions
Do I need to connect my computer for every task?
Plan according to the task’s actual sources and tools. A supplied document may support a draft without access to local files. Work that depends on your desktop environment has different requirements. Check the official computer guide and the task’s status. Do not provide local access merely because it appears in setup; first establish why the pilot needs it and how you will review the result.
What is the best first business assignment?
Our recommendation is a bounded internal draft using a small set of approved material. It should have a reviewer and a clear acceptance standard. A FAQ brief, meeting preparation note or source comparison can fit that description. Choose work that matters to your team, because the point is to learn whether the process is useful under ordinary conditions, not just to produce a successful demonstration.
Can the first task be website publishing?
It can be considered only after the relevant publishing access, permissions and review process are established. For a first test, preparing a draft is usually easier to assess and correct. Keep the content result separate from the publishing step. This guide’s FAQ example does not assume a WordPress connection or authorise any external changes. The website editor remains responsible for the final page.
Move from setup to better instructions
The strongest first task has a defined result, current sources, explicit boundaries and a reviewer who can check it. Save the improved brief after testing so future work begins from what you have learned. The next article provides original ChatGPT Dots prompt examples and explains how to adapt them. A useful AI workflow begins with a business process people can understand; clearer instructions then make that process easier to support and hand over.
