Build a repeatable AI visibility observation sheet

AI answers vary from one check to the next. A simple, consistent spreadsheet turns screenshots into a trend you can report honestly.

AI assistants give different answers to the same question. That makes single screenshots unreliable and makes many AI visibility reports hard to trust.

A simple observation sheet solves much of the problem. It turns scattered checks into a consistent record that can show change over time.

This article shows how to build one in a spreadsheet, using official guidance on how AI search features work.

Why a repeatable sheet matters

Google’s documentation on AI features says AI Overviews and AI Mode may use different models and techniques, so the responses and links shown can differ.

It also says AI Overviews are shown only when its systems judge them additive to classic Search, and as a result often do not appear.

Under those conditions, one check proves little. A sheet that repeats the same checks, in the same way, gives you a pattern instead of an anecdote.

Step 1: Choose 20 to 40 questions

Pick questions your buyers really ask. Use sales call notes, site search terms, Search Console queries and customer service logs.

Mix three kinds:

• Category questions, such as “best payroll software for small charities”.
• Problem questions, such as “how to reduce late invoice payments”.
• Brand questions, such as “is [your brand] good for [use case]”.

Write each question down exactly. The same wording must be used every time.

Step 2: Fix the conditions

Small differences change answers. Record and hold constant:

• The platform and mode, such as Google AI Mode or ChatGPT search.
• The country and language.
• Whether you are signed in, and whether history is on.
• The device type.

Use a fresh or neutral session where you can. Note any change in conditions in the sheet.

Step 3: Set up the columns

One row per question, per platform, per check. A practical set of columns:

1. Date
2. Platform and mode
3. Question (exact wording)
4. AI answer shown? Yes or no.
5. Brand mentioned? Yes or no.
6. Our page cited? Record the URL.
7. Position among citations, such as first of five.
8. Competitors cited
9. Accuracy note for anything the answer gets wrong about you.
10. Checker initials

Columns 4, 5 and 6 matter most. Keep “mentioned” and “cited” separate, because a brand can be named without a link.

Step 4: Repeat on a schedule

Check monthly for most businesses. Fortnightly makes sense during a launch or a major content change.

Run each question more than once per session if possible, because answers can vary. Record the share of runs in which you appear, not a single result.

Keep the same checker where you can. If several people help, write a one-page instruction sheet so everyone records results the same way.

Step 5: Add the official data

Observation is only part of the picture. Add platform data alongside it.

Google introduced generative AI performance reports in Search Console in June 2026. Google’s help page explains they show impressions from AI Overviews and AI Mode, with breakdowns by page, country, device and date.

The reports do not show clicks or query data. Google says they became available worldwide from 31 August 2026.

Google also says traffic from its AI features is included in the main Performance report under the “Web” search type. You cannot separate it there.

For other platforms, check analytics for referrals from AI services. Some visits will arrive without clear referral data, so treat these numbers as a minimum.

Step 6: Check access, not just answers

If a page never appears, first confirm it can be reached.

For Google, the AI features guidance says pages must be indexed and eligible for a snippet. Check this in Search Console before anything else.

For ChatGPT, OpenAI’s crawler documentation says OAI-SearchBot is used to show sites in search results. If your robots.txt blocks it, your pages will not be shown in ChatGPT search answers. OpenAI says robots.txt changes can take about 24 hours to take effect.

Ask your web team to confirm what is allowed. Blocking a training crawler, GPTBot, is a separate decision from blocking the search crawler.

Step 7: Report trends, not rankings

Summarise the sheet in three lines each month:

• Presence rate: the share of question-runs where your brand was cited.
• Accuracy issues: any incorrect claims about you, and who owns fixing the source.
• Competitor pattern: which rivals appear most often, and for which questions.

Avoid presenting it as a ranking. Answers vary too much for that.

An illustrative example

The following example is hypothetical.

A Dutch HR consultancy tracks 30 questions monthly on two platforms. In month one, it is cited in 4 of 60 question-runs. By month four, after improving three service pages, it is cited in 11.

The team reports this as an observed change in a small sample, not proof of cause. It also notes that impressions in Search Console’s AI report rose over the same period.

What the sheet will not do

It will not guarantee inclusion. No platform offers that.

It will not capture every variation, or every user’s personalised answer. It gives a consistent, honest sample, which is more than most teams have today.