AnswersIf you already do SEO

GEO vs SEO: what carries over, and what an SEO team adds

Generative engine optimisation (optimization, in US spelling), or GEO, starts from SEO. The assistants that search the web while they answer still need pages they can crawl, index and read, and Google says that for its own AI features the work is still SEO. What changes is the result and how it is measured: a written answer that names some brands and not others, a different answer on the next asking, and a picture of you drawn partly from pages you do not own.

Is GEO the same as SEO?

For Google Search, Google's own answer is that GEO is still SEO: optimising for its generative AI features is optimising for search. The work reaches past that in two places: other assistants, such as ChatGPT and Claude, run search crawlers of their own, and the result is a written answer that names some brands and recommends one, not a position in a list of links.

Google's guide to optimising for generative AI search takes the question directly: “From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” For AI Overviews and AI Mode, the work it describes is crawl access, indexing and what is on a page when it is fetched, which an SEO team already manages.

Two things reach past Google Search. OpenAI and Anthropic each run a search crawler of their own, OAI-SearchBot and Claude-SearchBot, and each is allowed or refused by its own name, so being readable by Google does not make a site readable by them. And Google says its AI features “can show what's being said about products and services across the web, including in blogs, videos, and forum discussions”, so what other pages say about a brand reaches the answer, and part of the work moves to pages the team does not own.

The vocabulary, GEO and the other acronyms in circulation, is on what generative engine optimisation is, and answer engine optimisation, or AEO, names the same work. This page is about what the difference means for a team that already does SEO.

Does SEO still matter for AI search?

SEO still matters for AI search, because the assistants that search the web while they answer rely on pages they are allowed to crawl and able to read. Google says the best practices for SEO remain relevant for AI Overviews and AI Mode, OpenAI says a site that blocks its search crawler will not be shown in ChatGPT search answers apart from navigational links, and Anthropic says blocking its search crawler may reduce a site's visibility in its search results.

Google's guidance for its AI features says the best practices for SEO “remain relevant” for AI Overviews and AI Mode, and that to be shown as a supporting link a page “must be indexed and eligible to be shown in Google Search with a snippet”. Its newer guide names one setting to check: a site “must be included in Search generative AI features in Search Console”. That setting is on by default, and set to exclude, it keeps a site's links and content out of AI Overviews and AI Mode. Delphi does not read Google's AI Overviews or AI Mode.

OpenAI says sites that opt out of its search crawler, OAI-SearchBot, “will not be shown in ChatGPT search answers”, though they can still appear as navigational links. Anthropic says blocking Claude-SearchBot “may reduce your site's visibility and accuracy in user search results”.

Google and OpenAI both name the network as well as the file. Google lists allowing crawling “in robots.txt, and by any CDN or hosting infrastructure”, and OpenAI recommends allowing its crawler in robots.txt “and allowing requests from our published IP ranges”. So a crawler can be allowed by the robots file and refused at the firewall, and both belong in the same review. The names to allow are on which AI crawlers to allow.

So the SEO team's existing work comes first. A refused crawler, a page that drops out of the index, or a site excluded in that Search Console setting keeps your own pages out of the answers that depend on it, and those answers can then describe you only from what other sites say.

What changes in an SEO team's reporting?

For AI answers, an SEO team reports rates rather than a rank: how often each assistant names the brand and how often it recommends it, one line per assistant, because the assistants rarely agree on which company they name most often.

There is no single reading to report. In our own study, of the 166 questions where at least three assistants named a company, every one of those assistants named the same company most often on 11. Of the companies named anywhere in the study, 75.7% were named by exactly one assistant. So a reading of one assistant is a poor stand-in for the rest. The study's most recent measurement ran on 24 August 2026, and every figure and its method are on the study.

Each line holds two rates rather than a position: how often the brand is named, and how often it is the one recommended, which mentioned versus recommended separates. Named often and recommended rarely is a different problem from being named rarely, and reading the two rates together tells them apart.

And each rate comes from asking more than once. An assistant asked the same question more than once inside one measurement named an identical set of companies 21.5% of the time, so one check is one sample. When a change counts as a result is on how many questions a measurement needs, and why the unit, the target and the measurement all differ from SEO is argued on what generative engine optimisation is.

What should an SEO team keep doing?

An SEO team should keep everything that makes a site reachable and readable: crawl access in robots.txt, at the CDN and in Google's Search Console setting for its AI features, indexable pages, structured data that matches the visible text, important facts in plain text, and original content rather than a restatement of what is already published. Google lists most of these in its own guidance for its AI features.

Crawl access, in the file, at the edge and in Search Console. The same review as before, with more names in it. A training crawler and a search crawler are separate names with separate consequences, and OpenAI states that each of its settings “is independent of the others”. Google adds a third place to check: the Search generative AI setting in Search Console, which includes a site by default.
Indexable pages. For Google's AI features a page has to be indexed and eligible for a snippet before it can be a supporting link, so an indexing problem carries straight over.
Structured data that matches the page. Google asks for structured data that matches the visible text on the page, which is the same rule its structured data policies apply across Search. Its newer guide adds that structured data “isn't required for generative AI search” and recommends keeping it for rich results, so it stays for the same reason as before.
Important facts in text. Google also asks for important content to be available in textual form. A price, a specification or a stock position that exists only in an image is a fact the page does not state.
Original content. Google's guide asks for “non-commodity content”, and warns that creating separate content for every variation of a query, done primarily to manipulate its AI responses, violates its scaled content abuse spam policy.

None of this needs a new team. It is work the SEO programme already does, with more crawler names in the review and one more setting to check.

What should an SEO team add for AI answers?

An SEO team should add four things: a fixed set of buyer questions written as sentences, readings of those questions repeated across each assistant buyers use, a check of what the answers say about the brand against its own pages, and a list of the outside sites the answers draw on, sorted by who can act on each.

Questions alongside keywords. Write down the questions a buyer puts to an assistant, in full sentences with their constraints, and keep the set fixed so one reading can be compared with the next. Delphi pins every measurement to a version of the question set, and the trend view says so where a comparison crosses a change to it. It does not estimate how many people ask each question: the set is a designed sample of buyer intent, not a volume table.
Repeated readings, across assistants. Put each question to each assistant more than once, and report how often you are named and how often you are the recommendation. Delphi puts each question to ChatGPT, Gemini, Claude, Perplexity and Grok more than once in every measurement, and publishes how far the assistants disagree with each other as a confidence band beside the Delphi Index, its overall score.
A check of what the answers say about you. Read the answers for what they state about your price, availability and product, and compare each statement with your own page. Delphi quotes both sides where an answer differs from your home page and does not rule on which is right, because a buyer reading the answer never opens your page, so the sale is lost the same way whichever side is right. When an answer contradicts your page covers what to do next.
The outside sites, and who can act on each. Where an answer shows its sources, list them for the questions a rival wins, then sort them by who can act: your own site, publications and reviewers, communities, and retailers. Delphi groups the sources it can identify the same way. A retailer listing is influenced through the brand's own account team at that retailer. Publications and communities are earned, by correcting an error, briefing a writer or answering in a community as yourself, and none of it puts the page under your control.

The first three sit naturally with an SEO team. The fourth needs whoever in the business already deals with those sites.

How do you tell whether SEO work is also moving the answers?

To tell whether SEO work is also moving AI answers, ask the same fixed set of buyer questions of the same assistants before and after the work, each more than once, and count only a change larger than the margin of error. Read three things separately: whether assistants can read the site, whether they name the brand, and whether they recommend it.

Read three things apart: whether the assistants can read the site, whether they name you, and whether they recommend you. The technical half of SEO work shows up in the first, which is a precondition rather than the payoff: a site can become fully readable and be named no more often, because the answer also depends on what other pages say. Ask the same questions of the same assistants before and after the work, and check whether a change is real before reporting it as a result.