Does this tell me my share of AI answers?

No. A handful of prompts checked by hand is an observation, not a market share, and a before and after does not prove the change caused the difference. Nothing here promises a citation or a ranking.

Does the product ask ChatGPT or other engines for me?

No. Checks are filed by a person, so no check calls a model or costs credits. Automated prompt runs are not offered.

Can an assistant read it?

Yes, through MCP (ai_visibility and ai_visibility_experiment), read-only and with the same permissions as the screen. It cannot file a check or conclude an experiment.

AI visibility: how your site shows up in AI answers, and testing a page change

Read four separate lanes (Google AI impressions, Bing AI citations, visits from AI assistants and form leads from their tagged links), and run small content experiments you check by hand before and after a change.

Steps

Open Analytics → AI visibility. Impressions and citations come from the Search Console and Bing exports you import on the Search and AI tab; without an import those lanes say there is no data, not zero.
Visits sent by AI assistants are not measured yet: the Google Analytics connection reads site totals, not visits by source. The lane says so instead of showing a number.
Form leads from AI assistants count accepted form submissions whose link carried an assistant's utm_source, such as chatgpt.com. It is a floor: assistants that do not tag their links are not seen. It is counted for the whole workspace, by language and country where the contact record has them.
The four lanes are never added together. An impression or a citation is not a visit, and a visit is not a lead. Visits from AI crawlers are not counted at all.
To test a change, write an experiment: the page, what changes, up to 10 prompts, the engines you will ask them in, a window on each side and a ceiling on checks (at most 120).
Before the change, ask each prompt in each engine yourself and file what the answer did: cited the page, mentioned you without a link, or not there. Start the experiment on the day the change went live, then file checks again after it.
Every check counts against the ceiling. The same prompt, engine and day filed twice counts once. When the ceiling is reached, conclude the experiment with a decision: expand, fix or stop.

Questions

Does this tell me my share of AI answers?: No. A handful of prompts checked by hand is an observation, not a market share, and a before and after does not prove the change caused the difference. Nothing here promises a citation or a ranking.
Does the product ask ChatGPT or other engines for me?: No. Checks are filed by a person, so no check calls a model or costs credits. Automated prompt runs are not offered.
Can an assistant read it?: Yes, through MCP (ai_visibility and ai_visibility_experiment), read-only and with the same permissions as the screen. It cannot file a check or conclude an experiment.

Updated 2026-10-25 · llms-full.txt