From open tabs to a decision you can defend

Define what would change your decision

Write a decision rule before asking for a recommendation. For example: proceed only if the workflow supports the needed source types, produces inspectable citations, and can be tested within a fixed time budget. Product documentation can establish a documented capability; it cannot prove that your team will save time. Keep those two kinds of evidence in separate columns.

Before you begin: connect the browser you actually mean

OpenAI documents a browser extension that can use open tabs and provide side chat. In the desktop setup, open Settings → Computer Use, choose your browser, and follow its installation flow. That route points you to the appropriate extension. Use the profile where you installed it. Availability and controls depend on your account, app version, and workspace settings.

This guide uses a public-page practice task. The cover is AI-generated editorial imagery; the five step graphics are conceptual illustrations, not screenshots of an actual ChatGPT session. Allow about 20 minutes for a first attempt; that is a suggested practice time, not a measured result.

Step 1 — Open side chat beside one public page

Open one of your sources and select ChatGPT from the browser toolbar or Extensions menu. Use the tab mention or page-context control your version provides. Begin in Chat for reading where available. If the panel cannot see the page, confirm the correct profile and connection before continuing.

Read only the page I selected. Tell me its title and URL, what information you can actually access, and anything you cannot verify. Do not change the page.

Step 2 — Name the three tabs and the decision

Open a public product page, official documentation, and an independent analysis of the same workflow. Select or mention each tab explicitly. Replace the placeholders in this prompt with the actual page titles or URLs. A list of links is not proof that ChatGPT has read every page; ask it to identify any failed or partial reads.

Use only these three tabs: [title and URL 1], [title and URL 2], [title and URL 3]. My task: Assess whether a small team should pilot an AI-assisted research workflow. Compare one product page, its documentation, and an independent analysis; separate capability claims from evidence of customer outcomes. Create a comparison with one row per option, the source URL for each important claim, and a separate column for missing information. Use “not verified” when a page does not establish a detail. Do not start trials, buy, register, or send messages.

Step 3 — Check the answer against the original pages

Open the source behind each decision-changing claim. Separate a visible statement from AI’s interpretation. For prices, check currency, billing period, paid-seat count, minimum seats, required add-ons, and any tax or fee exclusions. For events, check date, time zone, venue, and registration status. Do not convert a blank field into a favorable assumption.

Audit your comparison against the three source pages. For each important claim, show the supporting page and explain the calculation or inference. Mark contradictions and missing details. List what I must confirm before making the decision.

Worked example: a capability claim does not prove a time saving

Illustrative scenario. The sample pages below are invented to demonstrate the reasoning, not a report of a completed browser test.

Example decision note: The sources justify trying a bounded research task, but do not establish a team-wide time saving. The next question is whether our reviewer can accept the output faster while maintaining accuracy.

Create a decision record with five fields: decision, constraint, supporting source, unresolved issue, and next check. If an article cites a study, follow the original study and inspect the sample, task, date, and comparison before repeating its result.

Step 4 — Review the tab list, then group only those tabs

If browser actions and tab grouping are available in Work, select that mode and name Chrome explicitly. Reading pages and changing the browser are separate tasks. First ask for a preview of the three tab titles and URLs. After you check the list, request the grouping action below. Inspect the actual Chrome group afterward.

In Chrome, show me the exact three tab titles and URLs you would group as “Research workflow evidence”. Do not change the browser yet.

Group only those three reviewed tabs as “Research workflow evidence”. Leave the other tabs where they are. Do not close pages, navigate to checkout, register, or send anything. Report the group name and the tabs actually included.

If grouping is unavailable, create the group manually in Chrome and keep the comparison. A useful research result does not depend on this optional action. If the extension fails to connect, use OpenAI’s linked troubleshooting guidance; do not grant wider access just to complete a practice exercise.

Step 5 — Finish with a recommendation and one open question

Write a decision note under 150 words: the goal, the best-supported next step, the source links, the main tradeoff, and one unanswered question that could change the recommendation. Distinguish facts, calculations, and your judgment. If the evidence is insufficient, recommend the next check instead of choosing a winner.

Run the smallest useful pilot that can answer the remaining question. Measure total time from opening the sources to accepting the finished brief, including review and correction. Compare outputs against the same quality standard. Do not announce a productivity improvement because the first draft appeared quickly.

Try the same method on a single article

Select the article tab and ask for a short evidence review. This is useful when a headline makes a broad promise but the body describes a narrow example.

What is the main claim in this article? What evidence does the author present? Which statements are interpretation, which details remain unresolved, and what original source would I need to inspect before relying on the claim? Include the article URL.

Your finished-work checklist

• The comparison uses the three sources you selected.
• Decision-changing claims link back to an accessible page.
• Costs and dates use the correct assumptions.
• Missing information remains visible.
• The actual tab group matches the reviewed list, if grouping was used.
• The final note names a next step and one question.

Save the note with the date checked. When a price, policy, or event changes, revisit the original page before reusing the recommendation.

Put this into practice

Define the outcome, the evidence needed, and the smallest useful test before expanding the workflow.

Plan a pilot with a clear decision →