Search · 2026/09/24

You were named in the AI summary. The form still did not fire

At the end of September Aleyda Solís refreshed an AI-search checklist with a plain line: count citations and clicks as two things. Pew watched real US searches: with an AI summary up, people clicked classic results about half as often; they almost never clicked the source chips in the summary. A randomized test that hid the overview found outbound clicks down by about two-fifths when it was shown, with no lift in how easy people said it was to find an answer. The product will move the numbers. The direction stays: many people read the answer and leave.

Being named in the summary is still worth keeping, if that sentence matches the site. If the summary gets the city or the service wrong, that is more trouble than not being named, because the guest follows the wrong sentence. Whether you appear in the summary, and whether anyone submits a form, are two ledgers. Each month you read search clicks in Search Console. If generative impressions are there, you note which pages were mentioned. Whether a form succeeded is in GA4. Do not pull “ChatGPT named us” out as its own sales column, and do not use an industry click rate as the KPI. If a chat tool shows up as a source once or twice, that belongs in the monthly note. When forms do not arrive, open the landing page and see whether it shows a form or an email before you argue about the citation. Keep the two ledgers apart and the month will reconcile. The older note on what else to measure is the same four layers: search, visits, whether the summary is right, and whether anyone leaves their details.

Once a month the person in charge asks two chat tools the same three questions: what does this firm do, do they do systems work, and how do I contact them. If the answer is wrong, edit the matching passage on the site. Do not rewrite unrelated pages to pad the fix. The opening of the home page and the service page says what you do, where, and how to write. FAQs sit in visible HTML, not only inside an image in a collapsed widget. A case states the scope and the outcome. Do not force in a growth percentage nobody can check. We do not publish a phone number. If a summary invents one, edit the site and lock the contact line to the form or the email, so a later summary has a correct sentence to copy.

The first paragraph on the home page and the service page is what summaries tend to lift. The person editing the page writes the sentence a guest would search: what you do, where, and how to write. Do not stop at the studio name. We covered that in homepage first lines. The monthly sheet stays at two columns: search clicks, and successful forms. “We were mentioned” is a note, not a third sales column. When a brand query has impressions and no clicks, edit the title and the first paragraph that query lands on. See impressions without clicks. Structure: SEO/GEO.

Two scenes that match

Someone posts a screenshot in the group: the Google paragraph names the firm. The next week the person who reads the inbox still sees the same few forms. In the GA4 export we pulled for September, sessions from search were already few, and forms were fewer. ChatGPT and Gemini each showed up as a source once. That is not a surge. It means someone arrived from a chat. A monthly note is fine. Treating it as sales is not.

The more urgent miss is a wrong sentence. Someone asked why the phone does not connect. We do not publish a phone number. A wrong city is the same kind of error. When you are named, match that summary sentence to the passage on the site before you talk about whether traffic rose.

Check the summary against the site Structure

Background: @aleyda’s checklist (27 Sep 2026); Pew and the Agarwal/Sen-style public tests on overviews and clicks. Treat percentages as dated; read the primary papers.