How to win back traffic lost to Google AI Overviews

If you've watched your organic traffic graph slope downward through 2025 despite stable rankings, you're not alone. AI Overviews — Google's AI-generated answer block that sits above the 10 blue links — now appears on a steadily growing share of commercial searches, and they keep a meaningful share of clicks inside the SERP that used to flow to your site.

This isn't a panic post. AI Overviews aren't going away, but they aren't unwinnable either. The rules just changed. The brands showing up inside the Overview pulled ahead of the ones who hoped Google would put their result back where it used to be.

What an AI Overview actually is

An AI Overview is a paragraph-shaped synthesis that Google generates from a handful of sources and shows at the very top of the search results page. It usually:

• Spans 2–6 sentences of generated prose
• Cites 3–10 sources via small link cards below the answer
• Includes brand names directly in the answer text
• Sometimes embeds a comparison table or bullet list
• Appears on around a third of commercial-intent queries and rising¹

The key shift: the user reads the answer in the SERP itself. If your URL is one of the citation cards, you get a click. If your brand name is named in the answer text, you get attention even without the click. If neither — the search journey can end before your site is considered.

Why traffic dropped — the actual mechanism

It isn't just that the AI Overview occupies vertical space and pushes you down. Two compounding effects:

1. Zero-click answers. When the Overview fully answers the user's intent ("what is X", "how does Y work"), the user often doesn't scroll. Informational queries are hit hardest. SERP-watching data sources put the through-rate drop at 20–40% for informational intent.²
2. Selection bias toward "authoritative" sources. The model picks from a relatively small set of sources it considers trustworthy. Smaller publishers with weaker brand signals — even if they rank well organically — often don't make the citation list.

The second effect is the unfair one and the one we can do something about. The first is more structural; the response there is to optimize for inclusion in the Overview, not around it.

The seven changes that actually get you in the Overview

The order matters — the early items have the highest leverage.

1. Add Organization + WebSite + FAQPage schema, properly nested

JSON-LD is how Google's models understand "what is this page". A page with no schema is dramatically harder for the synthesis pass to bin correctly. The three to start with:

• Organization — gives the model your entity (brand name, logo, founding date, sameAs links to LinkedIn / Wikipedia)
• WebSite — declares the canonical search URL and brand
• FAQPage — turns your bulleted "common questions" into a structure the model can lift verbatim

If you're not sure where to start, our Schema Generator emits all three with one input. The validator at search.google.com/test/rich-results will confirm parse success.

2. Front-load the answer in question-style H2s

The model retrieves passages, not pages. A page structured as a series of question H2s, each with the answer in the first 1–2 sentences below, becomes a much richer source for the synthesis pass than a long narrative with the answer buried at the bottom.

A simple test: open one of your top pages. Can a reader who only reads the H2s and the first sentence below each H2 already get most of the value? If yes, the model can too. If no, restructure.

3. Cite your sources, in-text and visibly

Citation begets citation. Pages that themselves cite primary sources (with linked footnotes, not just "as Forbes reported") signal credibility to the synthesis pass. The model is, in a real sense, looking for sources that look like the sources it likes.

The footnote pattern at the bottom of this post is deliberate — every claim has a link to where the number came from. Adopt this on at least your tentpole posts.

4. Build mentions on Wikipedia-tier sites

A single mention in Wikipedia, a Substack writer's roundup, an industry-trade publication, or a high-trust forum thread can outweigh hundreds of generic backlinks for AI Overview inclusion. This is because the synthesis pass uses these higher-trust sources to disambiguate brand entities and decide which sites to cite at all.

Tactics:

• A Wikipedia article about your category (not your company) that links to your product as a representative example
• Inclusion in an annual "best X for Y" roundup from a known industry publication
• Founder-bylined contributions on Substack, Medium, or LinkedIn that link back to your tentpole posts

5. Unblock AI crawlers

Counterintuitive but bites a surprising number of sites: your robots.txt blocks GPTBot, ClaudeBot, Google-Extended, PerplexityBot, or all of the above. Check yours. If any of them are disallowed, you're invisible to the engine they belong to.

Our robots.txt Generator emits a permissive but audited rule set, or paste your current rules into the robots.txt Checker to see what's blocking what.

6. Publish an llms.txt

llms.txt is the new convention for telling AI crawlers what's worth indexing — think of it as sitemap.xml for the AI age. It's emerging fast as a standard and is checked by an increasing set of crawlers. Cost to add: ~10 minutes. Upside: you get a curated channel into the model's index.

Our llms.txt Generator produces a spec-compliant file from your site map. If you want the full spec, see our llms.txt tutorial.

7. Track AI Overview inclusion (not just rankings)

You can't optimize what you don't measure. Pick 20–30 prompts that match buyer intent for your category and check, weekly, whether your brand is mentioned in the Overview for each. Track:

• Mention rate — is your brand in the answer text?
• Citation rate — is your URL in the citation cards?
• Sentiment — positive, neutral, negative

Free version: do it manually in an incognito window. Tooled version: use FixAEO — that's our specific use case.

What about the deeper-intent click?

A separate question worth thinking about: when the AI Overview doesn't fully satisfy the user, where do they go? Increasingly, not to the citation cards but to one of three places:

• A different AI engine (Perplexity, ChatGPT, Claude) — they re-ask there for a more detailed answer
• A specific destination they already trust for that category (Reddit, a community forum, a known publisher)
• A "how to" or "comparison" page that the Overview deliberately punted on

The implication: your deeper, more tactical content (the "how to actually do this" or "comparison of options" posts) is now more important than your "what is X" posts. The "what is X" got swallowed by the Overview. The "how to" got more valuable because it's where the unsatisfied user lands next.

A two-month plan

If you do nothing else from this post:

Week	Action
1	Audit robots.txt, add Organization + WebSite schema, publish llms.txt
2	Refactor your top 5 pages to question-style H2 structure
3	Add FAQPage schema to those 5 pages
4	Set up weekly AI Overview tracking on 20 buyer prompts
5	Identify 3 high-trust mention targets, draft contribution pitches
6	Publish one tentpole tactical post with citation footnotes
7	Re-audit the 5 refactored pages against your tracking dashboard
8	Double down on whichever change moved the metric most

Two months in, you should be seeing measurable mention-rate gains on at least a quarter of the prompts you track. That's the leading indicator. Click-recovery follows mention rate by a quarter or so, in our experience.

The honest take

AI Overviews are a permanent fixture. Wishing them away doesn't help; the channel structurally rewards different signals than it used to, and the brands optimizing for those signals are pulling further ahead every month. The seven changes above are the cost of staying competitive. The good news is they're all under your control, none of them require permission from Google, and the ones with schema and llms.txt and robots.txt are durable infrastructure — you do them once and they pay back for years.

If you want to short-circuit the audit step, run a scan on your own site — we'll surface every one of these seven items, ranked by severity, in 3 minutes.

FAQ

What share of Google searches now show an AI Overview?

Roughly a third of commercial queries as of early 2026, with the share rising month over month. Informational queries trigger Overviews more often than transactional ones; navigational queries rarely.

Do AI Overviews count as zero-click results?

Often, yes. When the Overview fully answers the user's intent, the user typically doesn't scroll to the citation cards. That said, the citation cards do still receive meaningful click-through for users who want a deeper answer — so being cited inside the Overview is still worth optimizing for.

Can I opt my site out of AI Overviews?

Sort of. Adding Google-Extended: Disallow in robots.txt tells Google not to use your content for Bard and AI Overview generation, but it does not affect normal Google Search indexing. Most sites should leave it allowed — opting out means you can't be cited, which is worse than the alternative for almost everyone.

How long does it take to start appearing in Overviews after a change?

Schema and llms.txt changes are typically reflected within days. Citation-authority signals (Wikipedia, industry pubs) take weeks to months. Content restructuring effects show up at the speed of Google's next crawl + index pass — usually 1–2 weeks for active sites.

Does the AI Overview ever show only my brand?

It can on branded queries (someone searches your exact brand name). On generic queries it always names 2–5 brands. Single-brand inclusion is rare and usually means you dominate the topic; aim for being one of 2–3 named, not the only one.

What's the single highest-leverage change?

Adding FAQPage and Organization schema. Both take under an hour, both feed the model the structured information it needs to decide whether to include you, and both are usually missing.

Related reading

• How to get cited by Perplexity: a tactical playbook for 2026 — same problem, different engine
• Why ChatGPT doesn't recommend your brand — diagnostic walkthrough
• GEO vs AEO vs SEO: which acronym actually applies to your work? — terminology primer
• How to add llms.txt to your website in 10 minutes — the infrastructure piece

Footnotes

1. Coverage statistics here aggregate from multiple SERP-monitoring sources; ranges given are conservative midpoints from the public reporting in early 2026. Exact percentages drift week-to-week as Google adjusts the Overview triggering threshold. ↩

2. Click-through-rate impact varies wildly by query type. Informational queries see the largest drops; transactional and navigational queries are much less affected (the user still has to click to complete the action). ↩