AI Overviews Cut Position One CTR From 27% to 11%. The Recovery Playbook.

For fifteen years the deal was simple. Earn position one and roughly a quarter of searchers click you. SISTRIX measured what happens when Google puts an AI Overview above that result: click-through on position one drops from about 27 percent to about 11 percent.

Same ranking. Less than half the clicks. And that is the good case, where you still rank first.

Most agencies responded the way they always have: rewrite the title, tighten the meta description, add the keyword to the H1. We did too, at first. It did not work, and it is worth understanding why before you spend another quarter on it.

Why the meta rewrite stopped working

A title tag is an argument to a human scanning a list. Under an AI Overview, the human is not scanning a list. She is reading a paragraph that already answered her question, with three sources footnoted inside it. Your blue link sits below that paragraph, past the fold on mobile, competing with an answer that has already satisfied her.

The title never gets read. So optimizing it optimizes nothing.

What gets read is the answer. And the only way to be in the answer is to be one of the sources the engine chose to cite when it composed it.

Where the clicks went

Digital Applied measured the other side of the same shift: brands cited inside AI Overviews saw organic clicks rise around 35 percent. The traffic did not vanish. It moved from "the top result" to "the cited source," and those are increasingly different pages.

We see this in our own client data. For the questions that trigger an AI Overview in a client's market, the cited sources are rarely the business's homepage. They are a review site, a directory, a forum thread, or a specific page on a competitor's site that answers the exact question with a number in it.

The recovery playbook

Three moves, in this order. Each one builds on the last.

1. Get cited

Start by finding out which of your customers' questions already trigger an AI answer, and who is being cited. We run this as a scheduled measurement across ChatGPT, Perplexity, Gemini, and Google AI for every client, but you can do a rough version by hand: ask the engines the ten questions a customer asks before hiring you, in your city, and write down the sources.

Then look at what the cited pages have in common. In our data it is almost always the same three things: a direct answer in the first paragraph, a specific number or range, and a named place. "A typical shingle roof replacement in Hamilton County runs between X and Y and takes two to three days" gets cited. "We offer quality roofing solutions for all your needs" never will.

2. Publish structured data that matches the page

AI answers and Generative UI hydrate from structured data. If your page describes a service, mark it up as a Service. If it names the areas you cover, use areaServed. If it answers questions, mark them up as questions and answers.

Two rules that the tools never tell you:

• The schema must say the same thing as the visible text. Two nodes that contradict each other about the same business, one in the footer and one in a page block, are worse than no schema at all. We caught this on our own starter template: a page block emitted a second LocalBusiness with a different service area than the site-wide one. Engines read both and trust neither.
• Publish business identity once, site-wide. Per-page blocks render what the visitor sees. They do not re-declare who the business is.

3. Structure the evidence

The pages that win citations are built like evidence, not like brochures. A claim, then the proof, then the specifics.

• A cost guide with real ranges for your area, updated when your prices change.
• A page per real service, each answering "what, where, how long, what it costs, what it does not include."
• A FAQ written from the questions customers actually asked, in their words. Reviews and inbound messages are the best source of those words you will ever have.
• Reviews shown on the page, with the substantive ones first.

What to stop doing

• Stop reporting impressions as progress. A page that shows up under an AI Overview and gets no click is not a win.
• Stop mass-producing city pages. Ten pages with the city name swapped teach the engine that you have nothing specific to say about any of them.
• Stop treating FAQ schema as a rich result hack. Google stopped showing FAQ rich results for most sites in 2026. Keep the questions because people and engines read them, not because a snippet might appear.

How to know it is working

Track three numbers per client, monthly:

1. How many of the tracked questions cite the client, per engine.
2. How many leads arrived from an AI referral. They show up with a source of chatgpt.com, perplexity.ai, or gemini in the referrer, and we tag them automatically.
3. Clicks from search, measured from the Search Console API, not from a rank tracker. Rank trackers store rolling windows and will inflate any sum you make from them.

If the first number moves and the second follows within a month or two, the playbook is working. If the first number moves and nothing else does, the leads are reaching the site and dying there, and the next fix is response time, not content.