Something has changed in how people find information, and it has happened faster than most site owners have adjusted to. Roughly 60% of Google searches now end without a click to any website. In Google’s AI Mode, that figure reaches 93%.

When an AI Overview appears on a results page, organic click-through drops sharply — one analysis of 25 million impressions found it falling by around 61%.

Which raises an uncomfortable question: if fewer people click, what is ranking actually for? The answer is that being cited is becoming as valuable as being clicked, and the two are optimized for slightly differently.

How AI search picks sources

AI answers are assembled rather than ranked. The system takes a question, breaks it into related sub-questions, gathers information across several sources, and synthesizes an answer — citing the sources it drew from.

This process is sometimes called query fan-out, and it has a direct implication: a single excellent page on a topic is less likely to be cited than a site that covers the whole subject area thoroughly. The system is looking for sources that can answer several related questions, not one.

Topical depth, in other words, has become more valuable than individual page optimization.

What actually increases your chance of citation

Answer the question in the first two sentences

This is the single highest-value change most sites can make. A page that states its answer immediately is easy to extract from. One that spends eight hundred words on preamble before reaching the point is not.

It also serves human readers better, which is the useful thing about this particular optimization — you are not trading one audience for another.

Cover the whole topic, not one keyword

Because AI search expands beyond the original query, a site with ten interlinked pages covering a subject from different angles is far more likely to be drawn on than a site with one page targeting the main term.

Build clusters rather than isolated posts, and link them to each other so the relationship is explicit.

Include specifics that can be quoted

Numbers, dates, thresholds, named steps, comparison tables. Concrete facts are extractable in a way that general advice is not.

“Be consistent with publishing” cannot be cited. “One post a week sustained for a year outperforms five a week for two months” can.

Make your expertise visible

Experience, expertise, authoritativeness and trust are used to judge whether a source is worth citing. Practically that means named authors with real credentials, an About page describing who is behind the site, clear dates on content, and first-hand detail that demonstrates you have actually done the thing.

Anonymous, undated, generic content is the easiest thing for a system to skip.

Use clean structure and schema

Clear headings that match real questions, short paragraphs, lists where lists make sense, and proper structured data. None of this is new advice — it is simply more consequential now, because machines are parsing your page rather than only indexing it.

What still earns clicks

Not everything is being absorbed. Some queries still send people to sites, and knowing which ones should shape what you publish.

• Anything requiring a tool. An AI answer cannot run a calculation on your specific numbers. Calculators, generators and interactive content are unusually well protected.
• Decisions needing judgment. Comparisons where the right answer depends on the reader’s situation.
• Original data. If you have information that exists nowhere else, people come to the source.
• Ongoing relationships. Newsletters, communities and returning readers do not route through search at all.
• Transactional intent. People buying something still visit the site.

The pattern is clear enough: simple informational content is most exposed, and anything that does something is least exposed.

Track both, not one

Search Console shows impressions and clicks from Google, and impressions matter more than they used to — appearing in an AI answer builds recognition even without a visit.

What it does not show is whether AI systems are citing you. Checking manually — asking the assistants questions in your subject area and seeing who they reference — is crude but currently the most direct method available.

Frequently asked questions

Should I block AI crawlers?

Blocking prevents your content being used and also prevents you being cited or recommended. For most sites that trade is a poor one. Publishers with licensing arrangements or paywalled content are a different case.

Is traditional SEO dead?

No. AI Overviews currently appear on a minority of queries, and ranking well remains the main route into being cited. The fundamentals have not changed — their relative weighting has.

Does this hurt small sites more?

In one sense it helps them. Citation is based on how well a source answers a specific question, not on domain authority alone, so a genuinely expert small site can be cited alongside far larger ones — which was harder under pure link-based ranking.

How fast is this changing?

Quickly, and the figures in this article will move. Treat the direction as reliable and the specific percentages as a snapshot rather than a constant.

What to actually do

Answer questions directly and early. Build depth around subjects rather than scattering posts. Include facts specific enough to quote. Put a real name and real experience behind the site. Publish things that do something rather than only describe something.

All of which is what good content advice already said. The difference is that the cost of ignoring it has gone up considerably.

What this changes about publishing strategy

If simple informational content is the most exposed category, then a site built entirely from “what is X” and “how to do Y” articles is building on the part of the ground that is subsiding fastest.

That does not mean stop writing them. It means pairing them with something the answer cannot absorb.

• Attach a tool to informational posts. An article explaining break-even is easy to summarize. An article that ends with a calculator taking the reader’s own numbers is not.
• Publish comparisons that depend on the reader. “Which of these suits your situation” requires the reader to engage rather than receive an answer.
• Produce original figures. If a number exists only on your site, anyone citing it points at you.
• Build something that returns people directly. A newsletter or a genuinely useful resource creates visits that never touch a search engine.

The underlying shift is from being found to being worth returning to. Search was a discovery channel that happened to also deliver repeat traffic. It is becoming discovery only, and the repeat part has to come from somewhere else.

Why brand recognition now matters more

There is a second-order effect worth understanding. When an AI answer cites your site by name, some readers do not click — but they register the name. Cited repeatedly across a subject area, you become the source people eventually search for directly.

That makes impressions genuinely valuable rather than a consolation metric, and it makes consistency of naming worth attention. A site that presents itself the same way everywhere accumulates recognition faster than one that does not.

It also raises the value of being right. Sources that get cited and turn out to be wrong lose that position quickly, and in a system assembling answers from several places, accuracy is the thing being selected for.

A practical checklist

1. Rewrite your openings. Every post should answer its own title within two sentences. This is the highest-value change available and it takes an afternoon.
2. Group related posts into clusters and link them to each other explicitly.
3. Add a named author and a real About page. Anonymous content is the easiest thing to skip.
4. Date your content and update anything with figures in it.
5. Include at least one concrete, quotable specific in every post — a number, a threshold, a named step.
6. Check structured data is valid so machines can parse what your page contains.
7. Test what the assistants say about your subject area, and note who they cite.

None of that is exotic. It is the same advice good sites have followed for years, with the difference that the penalty for skipping it has become immediate rather than gradual.

What not to do about it

Two reactions are common and both make things worse. The first is producing more content faster on the theory that volume compensates for lower click-through. It does the opposite — thin, templated output is exactly what these systems filter out when selecting sources.

The second is stuffing pages with statements written for machines rather than readers. Content that reads as though it was assembled to be quoted tends to be poor at the thing it is imitating, and human engagement signals still matter.

The durable position is unglamorous: know something worth knowing, say it clearly and early, and be the source that turns out to be right.