AI search covers ChatGPT, Gemini, Perplexity and Google's AI Overviews. Here is how these systems pick sources and what that means for your site.
AI search is any search experience where a language model writes the answer instead of handing you a list of links. That covers ChatGPT, Gemini, Perplexity, Copilot and Google’s own AI Overviews and AI Mode. The user asks in full sentences, gets a synthesised answer, and sees a few citations if the system chooses to show them.
Why it behaves differently from a normal search
A classic search engine matches a query to documents and ranks them. An AI search system does something extra: it breaks the question into parts, retrieves passages for each part, then writes a single answer from all of them. That retrieval step usually runs on a conventional search index, which is the detail most people miss.
So the ranking fundamentals still apply. If a page cannot be crawled, is not indexed, or never surfaces for the underlying subqueries, it won’t be used. What changes is what happens after retrieval. The model is choosing sentences, not pages, and it is weighing whether your wording is clear enough to reuse.
Where the answers come from
Three sources feed a typical response. Live retrieval from a search index, which is how anything recent gets in. The model’s training data, which is why brands with years of coverage get mentioned without any citation. And whatever the user pasted or uploaded into the conversation.
That second source explains a lot of the frustration. A model can recommend a competitor with no link and no source, simply because it read more about them during training. You cannot fix that with a meta tag. You fix it by being written about more often in places the models read: trade press, review sites, forums, Reddit, YouTube transcripts and comparison roundups.
What actually moves the needle
- Be quotable. One idea per paragraph, the answer near the top, and headings phrased the way people ask. Models extract cleanly written passages and skip walls of qualification.
- Be specific. Prices, timescales, numbers, named products. Generic marketing copy gives a model nothing to cite.
- Be consistent everywhere. The same company description, service list and locations across your site, your listings, your profiles and your press coverage. Contradictions make a model hedge or pick someone else.
- Be technically reachable. Server-rendered content, sane robots rules, and a decision on whether to allow the AI crawlers. Blocking them protects nothing useful for most businesses and removes you from citation.
How much traffic is at stake
Less than the headlines suggest, for now. AI assistants send a small fraction of the referrals that Google does for most UK sites, often under 2% of sessions. But visitors arrive late in the decision, convert better than average, and the share grows every quarter. The bigger near-term effect is indirect: informational queries losing clicks to summaries while commercial queries stay intact.
Measure it properly before you react. Segment assistant referrers in analytics, track which prompts mention you, and watch branded search volume, which is where influence without a click shows up. We build that tracking into our AI search work so decisions come from numbers.
Common questions
Is AI search replacing Google?
No. Google still handles the overwhelming majority of searches, and it has folded AI into its own results rather than losing the query. What is changing is the share of questions answered without a click.
Do I need a separate strategy?
Not a separate one, an extended one. The technical foundations and content quality that win rankings also win citations. The additions are structure, specificity and off-site mentions.
Should I block AI crawlers?
Usually not. Blocking them keeps your content out of the answers people are already reading. Publishers with licensing deals are the exception.
Can I track where I appear?
Partly. Tools now run prompt sets against the major assistants and report which brands and domains get cited, giving you a trend line even though answers vary between runs.
For a short definition and related terms, see AI search in our SEO glossary.
