How to Show Up in Google AI Overviews & AI Mode

AI Overviews vs AI Mode: what you're actually optimizing for

Google now has two AI answer surfaces, and they behave similarly enough that you optimize for both at once. **AI Overviews** is the AI-written summary at the top of a normal results page, with a few cited links woven in. **AI Mode** is a separate, conversational search experience that goes deeper — Google describes it as able to "dive deeper into the web than a traditional search."

What ties them together — and the single most important idea in this guide — is *query fan-out*. Google confirms both surfaces use it: "AI Overviews and AI Mode may use a 'query fan-out' technique — issuing multiple related searches across subtopics and data sources — to develop a response." That one technique is why showing up in Google's AI answers is a different game from classic ranking, and why it has more in common with generative engine optimization than with chasing a keyword.

How query fan-out actually works

In classic search, your question runs as one query against one ranked list. With fan-out, Google does something different — in its own words, AI Mode "uses our query fan-out technique, breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf." The Deep Search variant pushes further: Google says it "can issue hundreds of searches, reason across disparate pieces of information, and create an expert-level fully-cited report in just minutes."

Practically, the pipeline runs in five stages: **decomposition** (break the question into sub-questions), **expansion** (generate related and adjacent queries), **parallel execution** (run them all against the index at once), **synthesis** (stitch the best passages into one answer), and **contextual results** (present it with citations). The key consequence: each sub-query competes independently and pulls a *distinct* pool of sources. You're not fighting for one ranking — you're fighting for inclusion across dozens of mini-searches happening simultaneously.

The eight kinds of sub-queries Google generates

Fan-out doesn't just reword your question — it synthesizes new sub-queries that branch off it. Understanding the types tells you what to cover:

• **Equivalent** — the same question, reworded ("best CRM for agencies" → "top agency CRM software").
• **Follow-up** — the natural next question a person asks after yours.
• **Generalization** — a broader version of the topic.
• **Specification** — a narrower, more specific version.
• **Canonicalization** — the standard or official phrasing of the concept.
• **Translation** — the same intent in another language.
• **Entailment** — something logically implied by your question.
• **Clarification** — resolving ambiguity in what you asked.

Why your #1 ranking no longer guarantees a citation

Here's the part that breaks most SEO instincts. Because each sub-query pulls its own sources, the page that ranks #1 for your head term may never appear in the AI answer — while a page that ranks nowhere near the top gets cited because it nailed one specific sub-question. The data backs this up bluntly.

And the click economics have shifted hard. Ahrefs analyzed 300,000 keywords and found that when an AI Overview is present, the position-one result sees a **58% lower clickthrough rate** (December 2025 data — up from a 34.5% drop a year earlier). A separate, broader read from Amsive found an average **~15.5% CTR decline across all keywords**. The two numbers aren't contradictory — they measure different scopes: the first is "what happens to #1 when an AIO shows up," the second is "the average hit across everything."

There's real upside hiding in here: brands that *do* get cited in AI Overviews tend to earn more clicks, not fewer — being inside the answer is the new prize. The job isn't to fight the AI box; it's to be one of the sources it's built from. To know whether you are, you have to measure AI visibility and AI citations directly, because rankings won't tell you.

Tactics that actually move the needle (with evidence)

Most "GEO advice" is folklore. The strongest evidence we have is the peer-reviewed GEO research paper (KDD 2024), which tested content tactics across 10,000 queries and measured the visibility lift inside generative answers. These are the tactics that won, in order:

1. 1**Add direct quotations** from relevant, authoritative sources — the top performer, around a **41% relative visibility improvement**.
2. 2**Add statistics and concrete numbers** — roughly a **33% improvement**, and the strongest tactic for subjective "is this a good answer" scoring.
3. 3**Write fluently and clearly** — about a **29% improvement**. Clean, readable prose is easier for a model to lift cleanly.
4. 4**Cite your sources** — adding references lifted visibility around **28%**. Pages that show their work get pulled into answers more.
5. 5**Use an authoritative, easy-to-understand tone** — more modest gains, but they compound with the above.

Operational moves that match how fan-out works

• **Answer the whole question cluster, not one keyword.** Because fan-out decomposes into related and follow-up sub-questions, a page that covers the comparison, the use-case and the "what about X" angles lands in more sub-query result pools than a single-answer page. This is the core move in answer engine optimization.
• **Build atomic, extractable chunks.** Question-then-answer blocks, clear lists, comparison tables and tight FAQs let a synthesizer lift a clean passage. Burying the answer in a 400-word wind-up means it never gets extracted.
• **Keep pages fresh.** AI answers favor recently updated content — AI tools have been found to cite pages roughly 26% fresher than classic search surfaces. A "last updated" pass with new data is real signal.
• **Build topical and entity authority.** Google leans toward focused authority — a specialist page tends to beat a generalist site on the same sub-query. Go deep on a topic rather than thin across many.
• **Earn presence off your own site.** Reddit and LinkedIn are among the most-cited domains in AI answers. Being discussed across the web — not just on your domain — raises your odds of being cited.

Structured data, schema, and the "no special markup" truth

This is where a lot of bad advice lives, so let's be precise. Google's guidance is direct: "You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add." There is no "AIO schema." Anyone selling you one is selling folklore.

But that doesn't make schema useless. Structured data still helps Google *understand and qualify* your page — for rich results, for snippets, for knowing what your page is about. And snippet-eligibility is the literal entry ticket to AI Overviews. So the right mental model is: schema is good hygiene that supports normal Search eligibility, which is what makes you eligible for AI answers. It is not a separate AIO lever.

What about llms.txt?

You'll see llms.txt recommended as an "AI SEO must-have." Be honest about what it is: a standard proposed by Jeremy Howard in September 2024 to give language models a curated map of your content. It makes *no claim* of adoption by Google AI Overviews, and Google explicitly says no AI text files are needed to appear in its AI features. It's low-effort to publish and may help other tools, but it is not a Google ranking factor and not a requirement. Treat it as an emerging, unproven-for-AIO experiment — not a fix.

Common mistakes that cost you AI citations

These are the avoidable ones — each maps to something we already covered, and each quietly keeps brands out of Google's AI answers.

• **Keyword stuffing.** The GEO study didn't just find this ineffective — it found it *backfired*, with roughly a **9% decrease** in generative-engine visibility. Classic keyword-density tactics actively hurt here.
• **Assuming #1 = cited.** It doesn't. With top-10/AIO overlap down in the 17–38% range, banking on your ranking to carry you into the answer is the single most expensive assumption.
• **Chasing special "AI files" or custom schema.** Google says none is required. Time spent here is time not spent making content quotable.
• **Blocking or de-snippeting pages.** `noindex`, `nosnippet` and robots blocks remove the exact eligibility AI Overviews depends on. Audit these before anything else.
• **Thin, single-answer pages.** A page that answers one narrow query misses every other sub-question fan-out generates. Depth and coverage win the parallel game.

How to measure whether it's working

You can't improve what you can't see, and rankings are now the wrong yardstick. As iPullRank argues, the move is to measure inclusion in AI answers and entity mentions instead of position. Concretely, that means four steps.

1. 1

   Track citations and share of voice

   Run a fixed set of category questions against the AI surfaces on a schedule and record whether you're named, how prominently, and how you stack up against competitors. Because AI answers drift, the trend matters more than any single snapshot.

2. 2

   Watch your first-party AI traffic

   When Google's systems read or cite your site, real AI crawlers hit your pages and real AI referrals land on them. That server-side data is ground truth — not a synthetic sample — and it tells you which pages are actually being consumed.

3. 3

   Verify the traffic is real

   AI-crawler user agents get spoofed. Confirming a hit genuinely came from a known AI engine (vs an impostor) keeps your measurement honest before you act on it.

4. 4

   Change one thing, re-measure

   Add stats to a page, tighten its FAQ, refresh its data — then re-run the prompt set. AI answers are non-deterministic, so the only reliable read is a controlled before/after.

This is exactly what SourceWatch is built for: it measures whether ChatGPT, Perplexity, Gemini and Claude cite your brand — your AI visibility and share of voice against competitors — and it captures the real, verified-vs-spoofed AI-crawler and AI-referral traffic landing on your site. There's also an MCP server so you can pull all of it straight into Claude Code while you work. If you're comparing options, see the best AI SEO tools.