Getting cited by AI without giving up SEO

Your page holds the top spot on Google. Ask ChatGPT the same question: it is nowhere to be found. The reverse happens too. The two rankings no longer line up.

GEO, generative engine optimization, tackles that gap. Getting cited by AI assistants, from ChatGPT to Google’s AI Overviews, without sacrificing the SEO that feeds them. One thread runs through it. GEO is not a new job. It is SEO, extended.

What is GEO?

GEO has one goal: getting a brand or a page into the answers AI assistants give. These surfaces no longer rank lists of links. They build an answer, and they pick their sources.

Most traffic still flows through Google. Visibility, on the other hand, is increasingly won inside the answer. This guide is eleven principles, in order: lay the foundations, produce material worth citing, get the brand circulating, then measure and invest. A four-step roadmap closes it.

Lay the foundations

Before content comes structure, and three questions. How SEO and GEO fit together. What base to build on. Whether AI can even read the site.

01One content system, many surfaces

SEO and GEO share the same groundwork: technical health, entities, E-E-A-T, clarity, authority. They do not share the same outcome. Ranking at the top of Google does not guarantee an AI citation. Getting cited by AI does not require visibility on Google.

The temptation is to run two tracks. “SEO content” on one side, “GEO content” on the other. Bad idea. The two cannibalize each other: same intents, two pages, diluted signals.

02Classic SEO is still the foundation

Technical SEO, architecture, internal linking, E-E-A-T, entities, structured data, crawlability: these still decide what gets retrieved. GEO adds a layer. Mentions, freshness, extractable formats, bot access. It does not replace the foundation.

Retrieval and citation are two different steps. Being retrieved is not being cited. Models pull far more pages than they display. Citation then comes down to four factors: how clear the passage is, perceived authority, the brand’s weight across the web, and freshness.

The goal is to extend SEO. Never switch it off to “do GEO.” The entity work and the speed work on this site are that foundation, not a side track: how to become a node, a graph a machine can read, and whether the page arrives in time to be fetched.

03If AI crawlers are blocked, nothing else matters

Before an AI can cite a page, it has to be able to read it. Plenty of sites make themselves invisible without knowing it. A line in robots.txt. An overzealous WAF. A “Block AI bots” CDN rule. A forgotten User-agent: * Disallow: /.

Every player runs three families of bots. Each one is handled differently.

Search bots first

They decide citation. A site that blocks OAI-SearchBot disappears from ChatGPT Search answers. It only survives there as a plain navigation link. Allow about 24 hours for a robots.txt change to take effect. Anthropic gives the same warning: blocking Claude-SearchBot can reduce visibility in Claude.

On-demand fetchers next

Their rules are fuzzier. Because the visit is triggered by a user, OpenAI warns that robots.txt may not apply to ChatGPT-User. Perplexity goes further: those rules generally do not apply to Perplexity-User. Only Anthropic states that Claude-User respects them. ChatGPT-User has no say in what gets included in Search. For these bots, the real blocking risk sits in the WAF and the CDN.

Training bots last

Blocking them is a legal decision, not a visibility one. Blocking GPTBot or ClaudeBot does not stop you from being cited in search. Google-Extended is a special case. It is not a crawler. It is a robots.txt token. Googlebot crawls as usual. The token only tells Google whether Gemini can train on the content and use it to answer inside its app. It has no effect on AI Overviews or AI Mode, which rely on Googlebot and the standard snippet controls: nosnippet, data-nosnippet, max-snippet, noindex.

Two settings to check this week

Since 15 September 2026, Cloudflare has replaced its “Block AI bots” switch with three categories: Search, Agent and Training. For new domains, Training and Agent are blocked by default on pages that show ads, and Agent covers bots like ChatGPT-User and Claude-User. The trickier part: blocking Training also blocks Googlebot, Applebot and BingBot.

On Google’s side, Search Console now offers a setting to opt out of generative AI features. Turn it on, and you lose every impression and every visit from AI Overviews and AI Mode.

Produce information worth citing

A readable site is not enough. It has to be given something to cite: new information, in the right place, kept up to date.

04Every piece must bring something the corpus does not have

A model recombines what it has already seen. It can come up with new phrasing. It cannot produce a real-world fact it has never read.

So these systems favor what looks original and lived: first-party data, product tests, interviews, protocols, internal numbers, expert takes. Generic content churned out by agents fills space in the short term. It does not last.

05AI recommendations run through editorial, not product pages

Ask an AI for “the best X” or “which brand for Y.” It leans on guides, comparisons and lists, often published by third parties. On that intent, product and category pages underperform.

Everything flips once the conversation is already about a specific brand. The product page takes over again. Choosing a brand and buying a product are two different intents.

The goal: an editorial magazine on your own site, plus placements in the “best X” listicles AI already cites. The catalog will not do the recommending.

06Freshness matters more in GEO than in classic SEO

Assistants, led by ChatGPT, prefer citing content that is recent or clearly updated. A strong older piece can still surface. A piece frozen for a year and a half starts with a handicap.

ChatGPT is the most sensitive to freshness. Citations are also far more short-lived than a Google ranking. From one month to the next, 40 to 60 percent of the domains cited change for the same question. Over six months, 70 to 90 percent (Profound, 2025). The “publish once, visible for ten years” asset only exists on the AI side if you maintain it.

Build the brand beyond your own site

An AI does not only read your site. It checks the footprint across the rest of the web.

07Brand beats aggressive link building

For an AI, being named often, in credible contexts, counts for more than piling up backlinks. YouTube mentions, web mentions, branded search volume, branded anchors: that is the authority graph models read.

Google is moving the same way. E-E-A-T, core updates, trust signals. Links are not dead. A young domain can still rank, and nothing justifies giving up on technical SEO and popularity.

08YouTube first, the open web second

Among off-site signals, YouTube is the one most correlated with a brand showing up in AI answers. Not only on Google. The correlation holds on ChatGPT too. YouTube plays a double role. It is a place where brands get mentioned, and a domain AI cites directly.

Reddit, the press, third-party listicles and expert forums play the same role, to a lesser degree.

09Organic no longer means just Google

Google remains the dominant channel by volume. Around it, YouTube, Reddit, ChatGPT, Perplexity, Gemini and AI Overviews largely draw on the same material.

So one piece of content can live several lives. A solid article gets cited by an AI, then repurposed for social. A long YouTube video gets transcribed for SEO. A product test feeds a product page, a guide and a video.

The goal: create proprietary information, then put it into circulation. No more “one asset, one channel.”

Measure and invest

What is left is knowing what to track, and where to put the money.

10Steer a share of voice, not an isolated ranking

A keyword ranking says less and less about real visibility. Visibility also plays out in AI Overviews, chatbots, YouTube and Reddit.

• SEO share of voice, across a basket of queries.
• AI share of voice, across a panel of prompts your customers actually ask.
• Upstream of both, branded search volume remains a good proxy for awareness.

Two quirks call for a rigorous method. Variability first: two ChatGPT answers to the same prompt share only about 21 percent of their cited domains on average (Parse, 2026). Fragmentation second: in a recent study, 78 to 85 percent of cited domains appeared on a single platform only.

On Google’s side, Search Console now offers a report dedicated to AI Overviews and AI Mode, open to every site since 31 August 2026. Impressions only, no separate clicks, no split between the two surfaces. It is still the only first-party measurement available.

11Organic content is still the cheapest asset, if you keep it fresh

A well-made article, video or guide costs little compared with its lifespan: years in SEO, a stream of AI citations for as long as it stays fresh. Stacking proprietary pieces eventually reshapes a market. Bottom-of-funnel paid, meanwhile, keeps getting more expensive.

The pairing that works: build the substance, then put paid budget behind the assets that already perform, to speed up mentions and branded search. Paid accelerates organic. It does not replace it.

Paid is even moving into the assistants themselves. Since February 2026, ChatGPT has shown sponsored ads below its answers, targeted to the conversation. Launched in the United States, they are rolling out market by market. The ad sits next to the answer. It does not buy the citation.

Four layers, in this order

Order matters. Producing content before AI can read the site is pointless.

1. 1.Open up access. Allow search bots (OAI-SearchBot, Claude-SearchBot, PerplexityBot) and keep Googlebot fully open. Check that the WAF and CDN let on-demand fetchers through (ChatGPT-User, Claude-User, Perplexity-User), especially with Cloudflare’s new settings. Make the training call separately (GPTBot, ClaudeBot, Google-Extended). Check the AI setting in Search Console.
2. 2.Produce proprietary information. Magazine, YouTube, interviews, internal data, tests: one thing the corpus does not have, in every piece. An extractable structure: a crisp definition, lists, tables, the answer at the top of each section.
3. 3.Get the brand circulating. YouTube mentions first, then the open web, Reddit, third-party listicles, the press. Being cited elsewhere outweighs stacking pages on your own site that nobody sees.
4. 4.Maintain and measure. Refresh content that is 12 to 36 months old. Track the prompt panel, AI and SEO share of voice, branded search and the Search Console AI report. Use paid as an accelerator, only on what is already solid.

The bottom line

Getting cited by AI means being readable by its crawlers, bringing information its corpus does not have, getting the name out (especially on YouTube and the open web), keeping the content fresh, and measuring a share of voice rather than a ranking.

None of this replaces SEO. All of it extends it. When a customer asks an AI the question, someone will get cited. The job is making sure it is you.

On this site

• SEO is now a distribution game. Why the click is no longer the metric, and why mentions elsewhere decide the citation.
• Entity building for AI systems. How to become a node the systems can verify.
• A visual graph builder, not a form. Schema as a graph, not a list of fields.
• Speed as infrastructure. A page that does not arrive cannot be cited.