Notes · Opinion 004 · Search and AI

Sites are now built to be cited, not read

This summer, Google made three changes. Together, they point to one fact: the visitor to your page may no longer be human. The bad news is that there is no metric for this yet.

Serhat Belen

Founder, BLN Global · · 8 min read

Short answer

Google now tells you how often your page appears in AI answers, but not how many clicks those appearances bring. During the same period, it removed FAQ rich results and urged sites to prepare for AI agents to crawl them. The practical result is clear. Build your page to be cited inside an answer, not clicked in search results. Use direct answers, a clear question-and-answer structure, a clean accessibility tree and near-zero layout shift.

Google made three separate announcements over the summer. One by one, each looks routine. Put them side by side and they point in the same direction.

Three events

7 May: FAQ rich results were removed

On 15 June, Google removed expandable Q&A boxes from search results. FAQPage It also took the documentation offline. Rankings didn't change. What you lost was the extra space in the results. The tactic sold for years as "add FAQ schema and take up twice the space in the SERP" ended in a day.

3 June: AI reports arrived in Search Console

Google now shows how often your page appears in generative AI features such as AI Overview and AI Mode. The separate report includes impressions, page, country, device and date.

What is missing from the list matters more than what is on it: no clicks, no click-through rate, no queries. Google says it will add more metrics later.

In Google's own words: Google Search Central, Google Search News, June 18, 2026. New reports begin at the 1-minute mark. The cover image comes from our server. YouTube only connects when you press play. No external requests are made before then.

June: Chrome released its agent-ready site toolkit

Chrome released a three-part set so AI agents can browse sites: a new Lighthouse Agentic Browsing audit category, the WebMCP draft standard that exposes site functions to agents, and DevTools tools that mimic agent behavior. The audits are informational for now. They do not affect the score.

Chrome’s reasoning is the interesting part here. It describes how agents see the page like this:

Agents use the accessibility tree derived from the DOM as their primary data model.

Chrome Developers, agent-ready site toolkit, June 2026

In the same document, the reason for lowering layout shift also changed: no longer "so you do not tap the wrong place", so the agent does not click the wrong place so.

This category is now really in Lighthouse. While writing this, we measured our own home page and a fifth column appeared on the top line of the report:

Top line of the Lighthouse report: Performance 99, Accessibility 100, Best Practices 100, SEO 100, and 3/3 in the new Agentic Browsing category.
blnglobal.business ana sayfası, 26 Temmuz 2026, mobil. Beşinci sütun Chrome M150 ile gelen Agentic Browsing category. For now, it is informational and does not affect the performance score. A 99 performance score is normal drift. Measure the same page twice in a row and it can move between 99 and 100.

What the three say together

Rich results are closing, AI visibility is being measured, and the browser wants you to prepare your site for machine visitors. The common point: the page’s job is increasingly being clicked stops being being cited as a source in an answer becomes.

This doesn’t mean SEO is dead. In fact, old and unglamorous work such as accessibility, clear heading structure and structured data now matters more. The target has changed. You’re formatting the page for the system that will summarize it, not only for the person reading it. Most of the time, both want the same thing. That’s the good news.

The bad news is in the metric. An impression count tells you very little. Appearing in an AI answer does not mean the reader visited your site. Most of the time, they did not. They already got the answer. Google gave you a new counter without showing how it relates to revenue. Managing a budget by this count is flying blind for now.

We tested this on our own site

A few hours before reading Chrome’s documentation, we measured a layout shift of 0,287 on our Arabic pages. Lighthouse showed the cause directly: the Arabic font loaded late and reflowed the page.

0,2870

Layout shift (CLS) on Arabic pages

How it was measured Font font-display:swap instead of optional we did it. We measured the live site with Lighthouse. The same pages’ performance score rose from 83-85 to 98-100.

We made the fix for human visitors because the page jumped as you read it. Later, we saw Chrome’s documentation justify the same metric as a way to stop agents from clicking the wrong thing. Good work serves both. That may be the most encouraging point in this article.

A separate note explains how we found the shift and why it appeared only in Arabic: the final 18 points on a static site.

What Google says, what the research says

You need to separate two sources here because they don’t say the same thing. One is a set of third-party studies on generative search optimization. The other is Google’s own guide, published in May.

Google’s guide directly rejects several common recommendations. You don’t need a separate machine-readable file, special markup or Markdown for AI. Google says these are ignored. Structured data isn’t required for generative search, and there’s no special schema to add. You don’t have to split content into small sections, and there’s no ideal page length. There is one technical requirement: the page must be indexed and eligible to appear in search results with a snippet.

Publishing content that people find original, interesting and useful will do more for your long-term presence in generative search than any other advice.

Google Search Central, optimization guide for generative AI, May 2026

Third-party studies use a very different language. They report that %44,2 of citations come from the first %30 of a document, %53,4 of cited pages contain fewer than 1.000 words, and pages with a question-and-answer structure receive about 2,3 times more citations.

They do not exactly conflict, but they get mixed up. What the studies found was relation: this is what cited pages look like. What Google says is requirement: you don’t have to do any of this. A well-written page already gives a clear answer near the top and doesn’t drag on. These studies are probably measuring quality, not a formula. The danger is turning correlation into a prescription and chopping up the page for AI.

What to do on Monday

You won't go wrong where the two overlap. In order:

  • Put a clear answer at the top of every important page. Do this for the reader, not for AI: the first paragraph should make the page clear. That is the only concrete point both sources agree on.
  • Keep the Q&A section. Don't add schema. FAQPage markup no longer produces anything in Google, and Google clearly says structured data is not required for generative search. Visible Q&A text still helps the reader.
  • Don't lengthen or split the page. Google says there is no ideal length and no need to split content into small pieces. Write as much as the subject needs.
  • Don't create a separate file for AI. Google ignores these. Part of the "AEO package" sold this year does exactly that.
  • Take accessibility seriously. The button name, link text, heading order. This is the exact tree the agent sees.
  • Bring layout shift down to zero. It has two reasons now.
  • Don’t rush WebMCP. The standard is still a draft, and audits do not affect your score yet. Track it, but do not budget for it today.

A warning

Be skeptical of anyone who sounds certain in this area, including this article. Our data is six weeks old, and Google reports are still open to a limited number of sites. People selling AI optimization packages are selling that uncertainty too.

Our view is simple: every item above would still be good practice without AI. Write a clear answer. Don’t pad the page. Make it accessible. Prevent layout shifts. All of this was true in 2015. The only change is that poor execution now costs more.

FAQ

Should you remove FAQ schema from your site?

No need. Google says unused structured data does not cause problems and does not need to be removed proactively. It produces nothing in Google, but other systems may read it. For new pages, though, adding it brings no Google benefit.

Does appearing in AI answers bring traffic?

Unknown, because Google does not share click data. We only have impressions. Being cited inside an answer may help brand awareness, but you cannot measure that right now. If someone says they can, ask how they measure it.

Should you rewrite your pages for AI?

No. You don't need to rewrite it. Edit it. Put a clear answer at the top of each page, turn headings into questions and split up long sections. If the copy is already good, you're halfway there.

Should you install WebMCP now?

No rush. It is a draft standard, behind a flag in Chrome, and Lighthouse audits do not count toward the score. If your site has functions such as reservation, ordering, and search, add it to your watchlist. If you run a marketing site, it is not on this year’s agenda.

Which one should a small business start with?

Clear answers and layout shift. Both fit into a day, both help visitors apart from AI, and both can be measured.

Sources

Related work