Search engine optimisation has spent two decades built on one assumption: rank higher, get more clicks. AI Overviews broke that assumption on a meaningful share of queries, and the SEO industry is still catching up to what actually replaced it. The short version — being ranked is no longer the same as being seen, and being cited inside the answer is now worth more than the blue link sitting beneath it.
The click-through collapse
On searches that trigger an AI Overview, organic click-through fell as low as 0.61% — a number that made explicit what a lot of publishers were already sensing in their analytics: the search summary was answering the question before anyone reached a website. Click-through has since partially recovered, to roughly 2.4% on AI-Overview queries, as Overviews matured and users learned when to expand past them. Recovered is not the same as restored. Both figures sit far below what the same ranking position earned before AI Overviews existed.
The practical read for anyone running a website: a page can hold position one and still see traffic drop, because position one no longer guarantees a click the way it used to. That's not a sign the page failed. It's a sign the reward for ranking changed shape, and measuring success purely by rank position now misses most of the story.
How an AI Overview decides what to show
An AI Overview isn't writing from nothing — it's synthesising an answer from pages it judges relevant, well-structured and trustworthy enough to draw from, then presenting that synthesis with a handful of sources attached. That means the inputs that always mattered for ranking — crawlability, page speed, clear heading structure, a page that actually answers the query directly — still decide whether you're even in the pool the Overview can pull from. Nothing about AI Overviews makes technical SEO optional; it makes it the entry ticket rather than the finish line.
Which searches trigger an Overview
Not every query gets one. Informational searches — how something works, what something costs, comparisons between options — trigger AI Overviews far more often than narrow transactional or navigational searches, where a user already knows exactly what they want and is looking for a specific page rather than an explanation. If your traffic leans heavily informational — guides, explainers, "how much does X cost" content — you're in the segment of search most affected by this shift, for better and worse. It's also the segment where being cited by name inside the answer does the most for a brand that isn't yet well known.
Being cited beats ranking below
If the AI Overview answers the query and your site is one of the sources it names or quotes, you get visibility and credibility even without a click — your brand appeared in the answer the user actually read. If your site instead ranks third in the blue links below that Overview, you're competing for a sliver of the 2.4% who scroll past. Those are not equivalent outcomes, and treating them as the same goal is where a lot of SEO strategy is currently stuck.
This reframes the objective for content built to rank. The question is no longer purely "can this beat the competition to position one" — it's "is this written in a way an AI system can lift cleanly into an answer, with our name attached." Those two goals overlap heavily but are not identical, and the difference is where the newer discipline of generative engine optimisation — GEO — actually lives.
What actually correlates with getting cited
The data here runs against a lot of instinctive SEO advice. Unlinked brand mentions — your business named in an article or forum thread with no hyperlink back to your site — correlate with AI citation far more strongly than backlinks do: 0.664 against 0.218. That's a large gap, and it says something SEO practitioners have been slow to accept: the link itself matters less to these systems than the fact that your name appears, in context, associated with the right topic, across enough independent sources.
That doesn't make backlinks worthless — they still carry weight for traditional ranking and for referral traffic. It means a strategy built entirely around acquiring links, while ignoring how often and how credibly your brand gets mentioned by name, is optimising for the weaker signal. Digital PR, genuine industry commentary, and being the source journalists and forums actually quote now do double duty: traditional authority, and AI-citation likelihood.
llms.txt does not help — Google has said so directly
A proposed standard called llms.txt — a plain-text file listing what an AI crawler should read on your site — got a lot of attention as a supposed lever for AI visibility. Google has confirmed directly that it does not affect rankings. Adding the file costs a few minutes and does no harm, so there's no reason not to have one — ours is at /llms.txt. But treating it as a strategy, or paying someone specifically to implement it as if it moves the needle, is paying for a file, not a result.
What to actually do about it
- Keep the technical and content foundation intact — crawlability, page speed and clear structure still decide whether you're eligible to be cited at all
- Write in a direct-answer structure — state the answer plainly near the top of the section, in the form an AI system can lift cleanly
- Build genuine entity signals — organisation schema, consistent naming, and a clear "who we are" that AI systems can match against other mentions of your brand
- Invest in brand mentions, not just links — commentary, quotes and citations in other people's content, even unlinked, given the 0.664 correlation above
- Track citation, not just rank — measuring only SERP position now misses the outcome that increasingly matters most
- Skip llms.txt as a strategy — add it if you like, for free, but don't budget for it as if it changes anything
The uncomfortable summary
Ranking well is still necessary — you can't be cited from a page Google can't find or trust. But it stopped being sufficient. The businesses adapting fastest are the ones measuring AI citation alongside rank, not instead of it.
The traditional foundation still has to be there
It's worth saying plainly, because it's easy to over-correct: none of this replaces conventional SEO. A site that AI crawlers can't read, that loads slowly, or that has no clear content structure will not get cited regardless of how many unlinked brand mentions it accumulates elsewhere. Generative engine optimisation sits on top of the same technical and content foundation that always mattered — it adds entity and mention-building work, it doesn't subtract the basics. Anyone pitching GEO as a replacement for technical SEO, rather than an addition to it, is selling a shortcut that doesn't exist.
Measuring the right things going forward
Search Console still tells you about rank and traditional click-through. It does not tell you whether ChatGPT, Perplexity, Gemini or Google's own AI Overviews are naming your business when someone asks a relevant question. That gap is why AI-citation tracking has become a second, separate line of reporting rather than a subset of the old one — and why a monthly report that only shows rankings is now telling you half the story on any query where an AI Overview is in play.
This is the part of SEO that changed fastest, and most agencies are still pricing it as an add-on rather than building it in from the start. We treat traditional SEO and generative engine optimisation as one discipline, not two.