Search stopped being a list of links for a growing share of queries. Answers now arrive assembled, with a handful of sources credited underneath, which raises an obvious question about what that means for SEO.
The short version is that it extends the discipline rather than replacing it. The work of earning those citations has acquired its own label, generative engine optimisation, and this covers what the data supports about it, what it does not, and where the honest gaps are.
Key Takeaways
- AI search extends SEO rather than replacing it, and the technical foundations still decide what is possible.
- Zero-click searches on Google rose from 56% to 69% in the twelve months after AI Overviews launched.
- There is no single AI search to optimise for, and platforms cite very different sources.
- Analysis of millions of citations found most cited sources are ones brands already control.
- Content that ranks well organically tends to perform well in AI Overviews, so SEO remains the foundation.
- Measurement is the weakest link, because a zero-click answer generates no referral session.
What Changed, and Why SEO Teams Noticed
The shift shows up clearly in click behaviour. Similarweb data indicates zero-click searches on Google grew from 56% to 69% in the twelve months following the launch of AI Overviews.
That is a structural change rather than a fluctuation. The same query that previously sent a click now frequently resolves on the results page.
AI referral traffic is still small in absolute terms, generally reported at around one percent of total site traffic. It is also the fastest-growing referral channel, and its users tend to arrive further through their decision process.
The commercially significant effect is harder to see. When a buyer forms a shortlist inside a conversation with an AI assistant, no session is recorded anywhere, and the influence happens before any measurable touchpoint exists.
There Is No Single AI Search
This is the finding that should reshape how agencies talk about AI visibility, and most of them do not mention it.
An analysis of 680 million citations across ChatGPT, Google AI Overviews and Perplexity found that only 11% of domains were cited by both ChatGPT and Perplexity. The platforms are not drawing on the same pool.
Citation frequency differs even more sharply. A 2026 study of over 34,000 AI responses found brand citation rates varying by a factor of 46 between platforms, with ChatGPT citing brands in a fraction of a percent of responses while Perplexity did so far more often.
So “optimize for AI search” is not one task. Perplexity is citation-forward and draws on live web results, AI Overviews lean heavily on existing organic ranking signals, and ChatGPT blends live retrieval with training data.
Pick your platforms deliberately rather than chasing all of them. Which ones matter depends on where your buyers actually are.
Most Cited Sources Are Ones You Control
Here is the finding that contradicts the common assumption. Yext analysed 6.8 million AI citations across ChatGPT, Gemini and Perplexity and found that first-party websites accounted for around 44% of citations and business listings for around 42%.
Taken together, roughly 86% of cited sources were brand-managed. That is the opposite of the received wisdom that AI visibility depends primarily on third-party coverage.
The practical implication is encouraging for smaller businesses. Your own site and your listings are the largest single lever, and both are within your control without a budget for digital PR.
Listings deserve particular attention because they are so often neglected. Consistent business information across directories and profiles is unglamorous work that appears to carry real weight.
Inconsistency is the specific problem. An old address on one directory and a former trading name on another give a retrieval system conflicting information about the same entity, and conflicting information is a reason to cite someone else.
Your SEO Foundations Still Decide What Is Possible
The comforting part is that none of this replaces the fundamentals. Content that already ranks well in organic search tends to perform well in AI Overviews, because the underlying signals overlap substantially.
Crawlability matters more, not less. A page an AI system cannot access cannot be cited, so technical fundamentals sit underneath everything else.
Structure appears to help. Clear headings, direct answers positioned near the top of a section, and unambiguous statements of fact are all easier for a system to extract than the same information buried in narrative.
Entity clarity is the newer emphasis. AI systems need to establish who you are, what you do, and why you are credible, which makes consistent naming and clear about pages more valuable than they used to be.
What Appears to Drive Citations
The evidence here is thinner than the confident advice suggests, but several patterns recur across studies.
Directness matters. Content that answers a question plainly, early and in a self-contained way is easier for a system to lift than the same answer distributed across several paragraphs of build-up.
Evidence appears to help. Statistics, named sources, dates and quotations give a retrieval system something concrete to attribute, and the original Princeton research into generative engine optimisation found citation-friendly formatting produced measurable visibility gains.
Third-party corroboration still counts, even though brand-managed sources dominate the citation pool. Being described consistently across independent sites strengthens the entity picture an AI system builds of you.
Freshness matters more on some platforms than others. Systems drawing on live web retrieval favour recently updated pages, while those leaning on training data respond more slowly to changes.
The Measurement Problem
This is where honest practitioners and confident salespeople diverge, and it deserves a plain statement.
Recent survey data suggests around 54% of organisations plan GEO activity while only around 23% measure it.
That gap exists because the measurement is genuinely difficult rather than because teams are lazy.
The core problem is structural. A zero-click answer produces no referral, so the most commercially valuable outcome, appearing on a buyer’s shortlist, leaves no trace in analytics.
Be sceptical of anyone quoting precise conversion multiples for AI traffic. Published figures vary enormously between studies, and the direction of travel is better evidenced than the magnitude.
What You Can Actually Measure
Several things are trackable and worth setting up now. AI platforms increasingly appear as identifiable referrers in analytics, so segmenting sessions from those sources gives you a real if partial number.
Citation tracking tools have matured. Several platforms now monitor whether your brand or URLs appear in AI responses to defined prompts, and share of voice against competitors is measurable that way.
Prompt-level testing costs nothing. Run the questions your buyers actually ask through the major assistants yourself, note who gets cited, and look at what those pages do differently.
Watch branded search and direct traffic as a proxy. Someone who encounters you in an AI answer and later searches your name is invisible as an AI referral, but the lift shows up in branded volume.
Check what the systems say about you as well as whether they mention you. An inaccurate description does more damage than no mention at all, and correcting the underlying sources is a different job from earning a citation.
Where to Start
Fix the foundations before chasing citations. Crawlability, page speed, clear structure, and accurate listings all sit upstream of anything AI-specific.
Then choose your platforms and establish a baseline. You cannot demonstrate improvement without knowing your current citation rate for the prompts that matter.
Working with Spicy Web, an SEO agency with dedicated generative search capability, the question worth asking is how they measure it.
An agency that acknowledges the measurement gap and explains how it works around it is more trustworthy than one presenting AI visibility as a solved problem.
Ask what they would do in the first ninety days. Baseline measurement, technical remediation, and listings consistency are a credible answer, while a promise of citations by a specific date is not.
Conclusion
AI search is a genuine extension of SEO and an early one. The evidence supports the direction clearly and supports the specifics far less, which is exactly the moment when confident claims should be treated carefully.
Do the parts that are well supported. Keep your technical foundations sound, control the sources you own, structure content so it can be extracted, and build measurement before you build expectations.
SEO and AI Search FAQs
Does AI search replace SEO? No. It extends it, and the same technical foundations of crawlability, structure, and authority underpin both traditional rankings and AI citations.
Which AI platforms should I focus on? It depends on where your buyers are. Perplexity is citation-forward, AI Overviews lean on existing organic rankings, and ChatGPT blends live retrieval with training data.
Do all AI platforms cite the same sources? No. Analysis of 680 million citations found only 11% of domains were cited by both ChatGPT and Perplexity.
What sources do AI systems cite most? Analysis of 6.8 million citations found first-party websites and business listings together accounted for roughly 86% of cited sources.
Has AI search reduced clicks to websites? Yes. Zero-click searches on Google rose from 56% to 69% in the year following the launch of AI Overviews.
How much traffic comes from AI platforms? Currently around one percent of total site traffic for most businesses, though it is the fastest-growing referral channel.
Can AI visibility be measured? Partially. Referral sessions and citation tracking are measurable, while zero-click influence on buyer shortlists is not.
Do I need a separate GEO strategy? Not separate so much as extended. Technical health, content structure, and listings accuracy serve both traditional and AI search.
What should I fix first? Crawlability, page structure, and the consistency of your business listings, since those are foundational and within your control.
Should I trust GEO conversion statistics? Treat them cautiously. Published conversion multiples for AI traffic vary widely between studies, so the direction is better evidenced than the magnitude.
