SEO for AI search: how to get cited in AI Overviews and answer engines
Search is turning into answers. What changes for websites when Google, ChatGPT and Perplexity summarise the web, and what still matters.
For twenty years, search meant ten blue links. Now a growing share of searches end with an answer written by a model: Google's AI Overviews, ChatGPT search, Perplexity and Copilot. The question every site owner asks me is whether SEO still matters. It matters more, but the target has moved from ranking a page to being the source an answer is built from.
What AI answers are built from
Answer engines still start with retrieval. They search an index, pull the most relevant passages and then write a summary that cites some of them. That means the old foundations decide whether you are even in the pool:
- Crawlable and indexed. If Google has not indexed the page, no AI feature built on Google's index can cite it. I have diagnosed stores where only one page in a hundred was indexed; no amount of content fixes that.
- Fast and rendered. Content that only appears after heavy JavaScript is easier to miss. Server-render or pre-render anything you want quoted.
- Correct canonicals and clean pagination, so the engine knows which version of a page is the real one.
Write passages, not just pages
Models quote passages. A page that wanders for 2,000 words before answering is hard to cite. A page where each section answers one question in its first two sentences is easy. In practice:
- Use headings that match how people actually ask, such as "How long does a DUI case take in Tennessee?"
- Answer directly under the heading, then add the detail, the exceptions and the evidence.
- Add a short FAQ block for the follow-up questions people ask next.
- Use specific numbers, places and names. Vague copy gives a model nothing worth citing.
Prove you are the right source
Answer engines favour sources that look trustworthy, which in practice means the same E-E-A-T signals Google has rewarded for years:
- Real author and staff pages with credentials.
- Organization or LocalBusiness schema, with consistent name, address and phone everywhere.
- Original material: your own photos, data, case results and process, which nobody else can supply.
For a Tennessee law firm, this looked like 26 city pages, practice-area hubs, attorney profiles and a library of more than 120 articles answering the urgent questions people search after an arrest or an accident. Impressions roughly doubled across three months.
Measure the right thing
When answers appear on the results page, some searchers never click. So clicks alone undercount your visibility. Watch these in Search Console:
- Impressions: are you being shown for more queries?
- Average position for the questions you wrote passages for.
- Query mix: longer, question-style queries usually mean you are being matched to intent, not just keywords.
Across the sites I manage, impressions have moved first and clicks have followed. An automotive manufacturer reached 930K at an average position of 5.2.
What not to do
- Do not mass-produce thin AI-written pages. They are easy to detect and add nothing a model cannot already write itself.
- Do not block the crawlers you want citations from in robots.txt by accident.
- Do not chase a single "AI ranking factor". The fundamentals compound; the tricks expire.
The takeaway
AI search rewards the sites that were doing SEO properly all along: indexable, fast, structured and specific. The difference now is that each section of your page has to stand on its own as the answer to a question.