Your weekly AI visibility review

Start with the comparison

Confirm the question set, location and collection surface before reviewing a trend. Note changes to the sample rather than hiding them in the total.

Use a separate row for ChatGPT, Gemini and Google AI Overviews. Record the collection method too. These are different experiences, with different evidence.

Fix a slot for the review, such as the same morning each week, and keep it short. A regular, modest review is easier to sustain than an occasional long one. Use the same checklist each time so the notes line up.

Decide who joins. One person can run it, but the owner or whoever answers enquiries can add context the figures lack.

Keep the scope list in the same document as the notes. When the scope changes, add a new dated line rather than editing the old one.

Check how much was captured

Read successful and failed collection counts together. A week with missing observations may not support the same comparison as a complete week.

If coverage changed sharply, investigate that first. Put the number of readable answers beside each percentage. A small sample can make a modest difference look dramatic.

Label each gap with its reason where you know it. A timeout, a blocked request and an empty reply point to different fixes. Gaps with no known reason get their own label too.

Compare coverage by assistant as well as in total. A drop on one assistant can hide behind a steady total.

Expect some searches to produce no AI Overview

Google AI Overviews behave differently from a chat assistant. Google says they are shown only when its systems judge them additive to classic Search, and that they often don't trigger.

So a check of an AI Overview can end three ways. One was shown and named you, one was shown and didn't, or none was shown at all. Keep the third as its own label.

Don't count a search with no AI Overview as an answer without a mention. It is closer to a missing capture, and it belongs beside your coverage figures.

• Google guidance on AI features

Read mentions with their context

Read the answer behind the figure. An inaccurate service claim can deserve attention even when the business was mentioned more often.

Compare equivalent periods. If you changed the questions or switched collection methods, label the break. Keep the earlier series rather than silently replacing it.

Look for prompts where the result changed since last week. A new mention and a lost mention are both worth opening. Read the sentence around your name, not just the fact that it was there.

Note the alternatives named beside you. If a new business starts to come up for your main prompts, record its name and the prompts involved.

Keep prompts that include your business name in a separate row. They tell you how you are described, not whether you are named while someone is choosing.

When a figure moves sharply

Check scope and coverage before proposing a business explanation. Leave the cause open when the records do not establish one.

Next, ask whether the assistant itself changed. A product update can change many answers on the same day. Note the date and look for the same shift across several prompts.

Only then read the answers to see what changed in substance. Avoid changing the website in a hurry on the strength of one week. Wait for a second review to see whether the move holds.

Write down what you checked and in what order. If the move turns out to come from the method, that note saves the next person the same search.

Check the pages behind citations

Open cited pages for the questions that matter most to your business. Check whether the page supports the particular claim in the answer.

Separate your own website from other domains. A link to your domain does not tell you whether the customer visited it or understood your offer.

Watch for new domains in the links, such as a directory or a news page you haven't seen before. Open them and check what they say about you. An old directory listing with wrong details is a finding you can act on.

Check whether your own pages are linked for the prompts that matter most. If a competitor's page or a directory is linked instead, note which page and what it covers that yours doesn't.

Review one important business fact

Review a detail that could affect customers this week. Confirm the actual offer before asking an owner to edit a page.

Save an incorrect answer with its date and sources. Identify the page you control that needs attention. Treat an unverified detail as a question to investigate.

Rotate the fact you check each week, so the main details each get a turn over a month. Keep a short list of the facts and the date each was last checked.

If the answer is wrong, start the correction from the saved record rather than from memory. The guide on reviewing an incorrect answer covers those steps.

Check the same fact on each assistant in scope, because one may be right while another is wrong.

Look at enquiries separately

Keep customer actions as another evidence stream. A change in answer counts does not establish why an enquiry arrived.

When no measurement source is connected, write not measured. Ask your team whether enquiries mention an AI tool, while recognising that memory is incomplete.

Keep each source's figure with its own date range. Analytics, server logs and enquiry records may not cover exactly the same days, and a weekly review should not blend them.

If a connected source stopped reporting part way through the week, mark the week as partial. A partial week set beside a full one looks like a fall that never happened.

Know where each tool counts AI visits

In Google Analytics, check the default channel called AI Assistant. Google's definition covers visits from sources such as ChatGPT, Gemini, Deepseek, Copilot or Grok. It leaves out Google's AI Overviews.

Google says visits from AI Overviews are counted within overall search traffic in Search Console. They sit in the Performance report under the Web search type.

So your weekly note needs two rows for AI visits, from two tools, with different scopes. Don't add them together, and don't read either one as a count of mentions.

Write which tool each figure came from and the dates it covers. If a tool wasn't checked this week, write not reviewed rather than leaving the row blank.

Google's newer guide also documents a Generative AI performance report in Search Console. Check which report your property provides and its coverage.

• Google Analytics channel definitions
• Google guidance on AI features
• Google, optimizing your website for generative AI features on Google Search

Leave a short review note

Record the finding, the evidence gap and the chosen task. A useful note makes next week's review easier without creating another reporting exercise.

Repeat the review on the same terms next week. A later change in answers does not establish that your work caused it. Record both observations without joining them.

If nothing changed, say so in a line. A note that says no change, with the coverage and date, is still a useful record.

Store the notes in one place in date order. A folder of weekly notes becomes the history you need when someone asks what changed and when.

Give the task an owner and a review date. Keep the original wording and the proposed replacement together. The website owner decides what to publish.

Write the task so someone else could pick it up. Name the page or listing, the change proposed, the evidence behind it and the person who decides.

Keep a short list of tasks that didn't make the cut. Next week you can choose from it rather than starting again.

Check last week's task first. Record whether it was done, is waiting on the owner or was dropped, and why.

Step back once a month

Once a month, put four weekly notes side by side. Patterns that are hard to spot week to week, such as a slow drift in coverage, become easier to see.

Use the monthly view to review the question set itself. Retire prompts that no longer reflect what you sell, and add any that customers have started asking. Record each change with its date.

It is also the moment to check that the people and tools behind the review haven't changed. A new team member or a new collection method belongs in the notes.

A worked weekly note

This note belongs to the repair business in this guide's example. It shows a sensible length and order for a weekly record.

The scope was the same twelve prompts on three assistants in one town. Coverage was thirty-two readable answers from thirty-six attempts, with four timeouts on one assistant.

Mentions went from nine to eleven, and both new mentions came on prompts about weekend repairs. One of those answers linked to an old page that still lists Sunday appointments.

Visits in the AI Assistant channel were low and steady, and the Search Console figure was not reviewed this week. The team logged two enquiries that mentioned an AI tool.

The task is to ask the owner to confirm the weekend policy, then propose a correction to the old page. The owner decides what to publish. Next week's review uses the same prompts.

The note doesn't claim the old page caused the new mentions. It records two observations and one task, ready for next week.

Sources and method

A Carqo review method. Provider statements link to the relevant documentation.

• Google guidance on AI features Link checked 4 November 2026
• Google Analytics channel definitions Link checked 4 November 2026
• Google, optimizing your website for generative AI features on Google Search Link checked 4 November 2026

Suggest a correction