If you have been told that AI search means starting over, you have been sold something. The overlap between ranking on Google and getting cited by ChatGPT, Perplexity, Gemini and AI Overviews is not partial. It is most of the job.
The short version: AI systems do not have a separate internet. They read the same pages, follow the same links, and lean on the same authority signals search engines have used for two decades. What changes is how the answer gets presented to the user — not what makes your page eligible to be part of it.
Below is what carries over directly, what does not, and the handful of things AI search genuinely asks for that classic SEO never did.
Why the overlap exists
It helps to be specific about what these systems actually are.
- Google AI Overviews and AI Mode are Google. Same crawler, same index, same core ranking systems. The generated answer sits on top of retrieval that has already happened. If a page cannot rank, it is not in the candidate set to be summarised.
- Perplexity runs live retrieval over a web index. It fetches pages at query time, reads them, and cites them. Crawlability and page speed are not background hygiene here — they are the entire entry condition.
- ChatGPT search retrieves from web indexes and its own crawler. Same requirement: your page has to be findable, fetchable and readable at the moment of the query.
- Everything else — the training data, the model's background sense of who you are — is built from the open web. Which is to say: from pages, links and mentions.
None of these systems invented a new way to find out that your business exists. They all inherited one.
The SEO work that transfers directly
If you are already doing these, you are already doing AI search optimisation. You do not need a second budget line for most of it.
- Crawlability and indexation. A page that Googlebot cannot reach is also a page GPTBot, PerplexityBot and ClaudeBot cannot reach. Broken redirects, orphaned pages, blocked directories and sitemap gaps cost you the same thing in both worlds — eligibility.
- Authority signals: links and mentions. This is the single biggest carryover and the least discussed. Retrieval systems have to decide which of a thousand pages answering a question is the one worth citing. They lean heavily on the same signals search engines do: who links to you, who references you, and whether the wider web treats you as a credible source on this topic. Editorial links built for Google rankings are doing double duty the day they go live.
- Topical depth across a site. One good page on a subject is weak evidence. Fifteen interlinked pages covering the subject properly is a pattern. Both Google's systems and LLM retrieval reward sites that demonstrably own a topic rather than sites that touched it once.
- Clear entity definition. Your business name, what you do, who you serve, where you operate — stated consistently on your site and matching what appears on your Google Business Profile, LinkedIn, directories and press. Inconsistency here has always hurt local SEO. It now also means models cannot form a stable picture of what your brand is.
- Structured data. Organization, Product, Service, FAQPage and Article schema were always about making a page machine-readable. That requirement did not soften when the machine started writing prose.
- Page speed and clean rendering. Content that only appears after JavaScript execution is a real risk with live-retrieval systems that often do not render it. This was already a Google problem. It is a harder problem now.
- Freshness. Updated pages outrank stale ones and get cited over stale ones. Same behaviour, two surfaces.
- Titles, headings and page structure. A clear H1, descriptive H2s and a logical hierarchy help a crawler understand the page. They also help a retrieval system find the specific passage that answers a specific question.
What does not transfer
Being honest about this is the difference between a strategy and a sales pitch.
- Position one does not guarantee a citation. Ranking makes you eligible. It does not make you chosen. Systems routinely cite the fourth result because its third paragraph answered the question more cleanly than the first result's whole page did.
- Keyword density and exact-match phrasing matter less. People ask AI systems full questions in natural language. Optimising for a two-word head term is optimising for a query shape that is losing share.
- Meta description tricks do nothing. There is no snippet to win.
- Thin programmatic pages actively hurt. Google has been de-indexing bulk low-value content. Retrieval systems are, if anything, harsher — a page with nothing specific in it has nothing to extract.
- Retrieval is not deterministic. Run the same prompt twice and you can get different sources. Anyone promising you a fixed position in an AI answer is describing something that does not exist.
What AI search asks for that SEO did not
Four things are genuinely new. They are additions to the SEO foundation, not replacements for it.
- Answer-first structure. Put the direct answer in the first paragraph under the heading, then expand. Content built to keep a reader scrolling toward a payoff is content a retrieval system abandons before it reaches the point.
- Passage-level self-containment. A retrieval system may pull one section of your page and nothing else. Each section has to make sense without the rest of the article around it. That means naming the subject rather than saying “it,” and restating context rather than assuming it.
- Third-party corroboration. You cannot optimise pages you do not own, and those are frequently the pages being cited. When someone asks a model for the best provider in your category, the answer is often assembled from listicles, forum threads, review sites and industry roundups — not from your website. Being present and accurately described in those places is a distinct piece of work.
- Extractable specifics. Numbers, dates, prices, named methods, defined terms. Vague positioning copy gives a model nothing to quote. “We help businesses grow” cannot be cited. “Managed link programmes from $1,200 per month, minimum three-month term” can.
How to check where you actually stand
Before changing anything, get a baseline:
- Confirm the AI crawlers are allowed. Check robots.txt for GPTBot, PerplexityBot, ClaudeBot, Google-Extended and Bingbot. Blocking them is common and usually accidental.
- Ask the systems directly. Run twenty questions a real prospect would ask, across ChatGPT, Perplexity, Gemini and Google AI Mode. Record who gets cited. That list of competitors is your actual competitive set.
- Check what the model thinks you are. Ask each system what your company does. If the description is wrong, thin or missing, the problem is entity clarity and third-party mentions, not your on-page copy.
- Cross-reference against your Google rankings. Where you rank well but are never cited, the issue is structure and extractability. Where you rank poorly and are never cited, the issue is authority — and that is the same fix it always was.
The practical conclusion
The businesses that will be visible in AI search are, to a first approximation, the businesses that were already doing SEO properly. Authority still comes from the same place. Clarity still matters more than cleverness. Links and mentions are still the currency.
What has changed is the cost of being vague and the cost of being invisible to a crawler. Both went up.
If you are already investing in search visibility, the question is not whether to start again. It is whether the work you are paying for is structured so that it counts twice.
Get a free AI visibility audit — we run your brand against the questions your buyers are actually asking and show you who is getting cited instead of you.
