How to Get Cited in Google AI Overviews

Most teams still measure AI search by whether they rank. That is now the wrong instrument.

Google shows an AI Overview on more than one in five US searches. When it appears, the answer often ends the session. SparkToro measured this using Similarweb clickstream data. Zero-click searches reached 68.01% of US queries between January and April 2026. In 2024 the figure was 60.45%.

The share of searches producing at least one click fell 9.51 percentage points over the same two years. That is a 22.9% decline in clicks available to everyone.

So the question changed. It is no longer whether you rank for a term. It is whether the model names you when it answers.

Ranking and citation have come apart

This is the finding that should reset your roadmap.

Ahrefs studied 863,000 keywords and roughly 4 million AI Overview URLs. In early 2026, only 38% of cited pages also ranked in the organic top 10. In mid-2025 that figure was 76%.

Read that twice. The overlap between ranking and being cited roughly halved in under a year.

The same analysis found about 31% of cited pages ranked beyond position 100, or did not rank at all. Pages nobody would call competitive are being quoted inside answers.

Two things follow. Your rank tracker no longer describes your AI visibility. And a page can earn citations without first winning a brutal SERP.

That second point is the opportunity. It is also why I stopped sequencing client roadmaps by search volume.

What a citation is actually worth

Being cited does not restore the traffic you had before AI Overviews. It recovers part of it.

Seer Interactive tracked 53 brands from January 2025 to February 2026. The sample covered 5.47 million queries. Where an AI Overview appeared and the page was cited, click-through rate was 2.1%. Where the page was not cited, it was 0.9%.

With no AI Overview present, the same pages earned 3.3%.

So a citation is worth roughly double an uncited placement. It still trails the old baseline. Anyone telling you AEO restores pre-2024 traffic is selling something.

What it does buy is presence at the moment of decision. On a considered B2B purchase, being named by the model beats a click from position seven.

Where AI Overviews actually show up

Coverage is wildly uneven, and that shapes where the work pays back.

Seer Interactive analysed 49,353 queries by type. AI Overviews appeared on 95% of comparison queries of the "X vs Y" kind. Question-format queries triggered them 86% of the time.

Informational queries sat at 36%. Transactional queries came in at 5%.

Your comparison and alternatives pages are therefore the highest-exposure surface you own. For most B2B sites, those pages were written once and never revisited.

That is usually the first thing I fix. The exposure is already there. The content simply is not built for extraction.

Five things that earn a citation

These are ordered by how quickly they move, based on the engagements I have run.

1. Say what you are in one sentence

A model needs to place you before it can recommend you. Your category and your buyer, plus the boundary of what you do not do.

Write it as a standalone sentence near the top of the page. Do not leave it implied across paragraphs of positioning language.

Vague positioning that a human infers from context is a dead end for a machine.

2. Make claims liftable

Answers are assembled from statements that survive removal from their surroundings. A sentence that only parses after the two before it will not be quoted.

Put the claim first, then the qualification. Short declarative sentences with concrete nouns get lifted.

3. Mark up the facts

Schema states facts in a form that needs no interpretation. Organization and Product markup carry most of the weight. FAQPage helps on question pages.

It is the cheapest work on this list and the most commonly skipped.

4. Get corroborated somewhere else

Models weight agreement across sources. Being described the same way on review platforms matters more than repeating it louder yourself.

This is the slowest item here and the hardest to fake.

5. Keep a dated trail

Freshness signals matter more in answers than in rankings. Visible publish and modified dates, plus real updates behind them.

How to measure whether any of it worked

You cannot improve what you do not score, and rank tracking will not tell you.

Build a fixed set of forty to eighty buyer questions. Run them across Google AI Overviews, ChatGPT, Perplexity and Gemini on a monthly cadence.

Score each result on three levels. Mentioned means your name appears. Cited means your page is linked or named as a source. Recommended means you are put forward as the answer.

Record which competitors appear beside you. That list is frequently not the one your sales team expects.

The set has to stay fixed. Change the questions between runs and you have destroyed your ability to compare anything.

I ran this programme on one B2B fintech engagement. It went from a zero baseline to more than 62,000 AI citations. Average position inside answers settled near five. The scoreboard is what made the roadmap arguable rather than a matter of taste.

Doing it by hand is miserable. I wire it to agents that read live data from Search Console, GA4 and SE Ranking. Measurement that is painful to repeat gets abandoned by month three.

What does not work

Adding schema to a page nothing links to. Structured data describes a page. It does not give that page authority.

Chasing volume. The questions feeding AI recommendations are long and conversational. Most show almost no volume in a keyword tool.

Publishing more of the same. If a model already understands your category and picks someone else, another article will not change its mind. Entity clarity and corroboration will.

Rand Fishkin, co-founder of SparkToro, put the strategic version of this plainly. He recommends "investing in brand awareness and influence on the platforms where your audience already spends time".

That is the uncomfortable part. A lot of AEO is not on-page work at all.

Where to start this week

Pull your comparison and alternatives pages first. Coverage is highest there, at 95% on "X vs Y" queries.

Write one standalone sentence on each that states category, buyer and boundary. Then check whether your top claims survive being read in isolation.

Run twenty buyer questions across the engines by hand. Write down who gets named. That takes an afternoon and it will tell you more than a month of rank reports.

If the answer comes back with a competitor every time, the gap is usually entity clarity, not content volume. That is the fastest thing on this list to fix.

Would you rather have the full baseline run for you? It is the first thing included in my answer engine optimization services. The audit call is free.