SEO for AI is the work of being the answer when the search engine writes the answer itself. Google AI Overviews, AI Mode, ChatGPT, and Perplexity do not show your buyer ten links. They compose a short response and name a few sources. This piece covers what carries over from classic SEO, what is genuinely new, and the order we run it in for client sites.
AI search is already sending buyers. Be citable.
Build the estimate to see the AI-visibility work we would ship for your site in the first thirty days. Sixty seconds, no opt-in.
The plain-English version
An AI answer is assembled, not ranked. The engine retrieves candidate pages, pulls out the clearest well-sourced statements, composes them into a response, and decides which sources deserve a citation. So SEO for AI is four jobs: be retrievable, be extractable, be worth quoting, and be safe to cite. A page can rank on page one and still fail three of those four jobs.
The good news is that the entry ticket is the SEO you should already be doing. Google is explicit that AI features in Search draw on the same index and the same helpful-content systems as classic results. If your site is crawlable, fast, and genuinely useful, you are eligible. The work on top of that is smaller than most agencies imply, and we walk through the eligibility layer in SEO foundations for Google AI search.
Why this moved from optional to urgent
The behavior shift is measurable. AI prompts average around 23 words against roughly 3 to 4 for a classic search, per HubSpot's AEO research, which means the engine reads intent precisely and hands back a short, named answer. A growing share of queries end without a classic click at all. We took apart what that does to traffic math in the zero-click piece. The demand did not disappear. It moved to whether the model names you, and that is winnable.
There is also a selection effect worth naming. A buyer who described their whole situation to an AI and got your name arrives pre-qualified. The pool is smaller than blue-link traffic. The intent per visitor is higher. For a business that sells considered work, that is a good trade.
What carries over from classic SEO
- Crawlability and indexing. If Googlebot cannot fetch and index the page, no AI feature can use it. Same robots.txt, same sitemaps, same canonical hygiene.
- Helpful content. The engines lean on the same quality systems. Thin, assembled-to-rank pages do not get quoted. Google's own helpful content guidance is still the bar.
- Real expertise. Named authors, first-hand experience, specific claims with sources. Models treat attribution as a trust proxy, which we covered in the E-E-A-T piece.
- Site speed and stability. Slow, erroring pages get crawled less and trusted less. Nothing about AI search forgives a broken site.
What is genuinely new
1. Answer blocks: write the quotable passage yourself
Every page that targets a question should open with a two-to-four-sentence answer a model could lift verbatim: the question in the heading, the direct answer immediately under it, no warm-up copy in between. Extraction engines take the clearest self-contained passage they can find. If you do not write it, they take someone else's. Run this on your top 20 to 30 pages before anything else. It is the highest-return hour of work in the whole discipline, and we go deeper on it in eight ways to perform in AI search.
2. A connected entity graph, not orphan schema
A lone JSON-LD block earns little. What engines follow is a graph: Organization schema naming the business, Person schema for the people with real credentials, Service or Product schema for what you sell, FAQ schema on your answer blocks, all cross-referencing each other and pointing at real external profiles. Consistency does the compounding: the same name, address, and category language everywhere the business appears. Inconsistent details fracture the entity and quietly weaken every mention.
3. Let the AI crawlers in, deliberately
ChatGPT search uses OAI-SearchBot. Perplexity uses PerplexityBot. Blocking them in robots.txt takes you out of those answers entirely, which is a legitimate choice for a publisher and a bad default for a business that wants customers. Decide per bot, on purpose. While you are in the file, ship an llms.txt: a short machine-readable summary of who you are, what you do, and where your best pages live. An hour of work, and it is reflected faster than deep content changes.
4. Distributed mentions
Engines corroborate. A claim that appears on your site and nowhere else is weak. The same facts on your site, your Google Business Profile, two directories, and one piece of earned coverage is strong. Complete the profiles you already have before chasing new ones. Agreement across sources is the trust signal a competitor cannot fake quickly.
The 30-day sequence we run
- Week 1: list the 15 questions your best buyers actually ask, in their words. Pull from sales calls, intake forms, and the People Also Ask box. These are the targets.
- Week 2: write or rewrite the answer blocks for those questions on your highest-value pages. Question in the heading, direct answer under it, detail below.
- Week 3: ship the entity graph (Organization, Person, Service, FAQ schema, cross-referenced) plus llms.txt, and verify the AI crawlers you want are allowed.
- Week 4: baseline your visibility. Ask ChatGPT, Perplexity, and Google AI Mode your 15 questions. Record who gets named. Repeat monthly and track movement.
How to measure it
Search Console now separates AI Mode and AI Overview impressions well enough to watch the trend, and we covered the workflow in measuring AI search visibility. For the non-Google engines, measurement is still manual: a monthly prompt panel of the questions you want to win, scored by whether you are named and linked. Crude, but it turns an anxiety into a number.
Watch your server logs too. OAI-SearchBot, PerplexityBot, and Google-Extended each identify themselves, so a monthly grep tells you which engines are reading you, how often, and which pages they pull. Rising crawl from an engine that has never cited you usually means you are in the retrieval pool but losing the citation step: the fix is almost always a sharper answer block on the pages being fetched, not more content. Referral traffic from the AI surfaces lands in GA4 like any other source, so tag it a channel and give it a conversion read alongside organic.
Want to know where you stand today? Run the estimator and we will read your market and your current AI visibility before any sales call. Or start with how we do this across the industries we serve, and the deeper walkthrough in how to show up in AI Overviews and AI Mode.