SEO optimizes a page to rank in a list of ten links. GEO (generative engine optimization) optimizes a page to be read, trusted and quoted by an AI that writes one answer. In SEO you can be seventh and still get a click. In GEO you are either cited or you do not exist.
A client told me his SEO was in great shape. Agency on retainer, reports every month, position three on the main query. I asked ChatGPT the question his own buyers ask. It named four companies. He was not one of them. “That can’t be right,” he said. It was right. It just was not the same game.
Below is the difference in one table, four things that do not change, the numbers from my own audits, and five decisions I would make this quarter if I ran a site that depends on search.
The short version
SEO and GEO share the foundation: crawlable site, clear entity, real content. They part ways on the unit of output. SEO competes for a position on a page. GEO competes for a sentence inside an answer, and the sentence is chosen on facts, structure and whether the bot could read your page at all.
Most sites that “do SEO” are not losing GEO on content. They are losing it on plumbing: blocked bots, pages rendered in the browser, no machine-readable facts. Those are fixed in weeks, not quarters.
SEO vs GEO in one table
Eight rows. Each one is a place where I have seen a team apply the SEO reflex to a GEO problem and wonder why nothing moved.
| SEO | GEO | |
|---|---|---|
| Goal | A position in a list of links | A citation inside one synthesized answer |
| Unit of output | Page in the top 10 | Sentence or fact attributed to you |
| Who reads the page | Googlebot, then a human | GPTBot, ClaudeBot, PerplexityBot, Google-Extended, then a model |
| What wins | Relevance, authority, links, intent match | Verifiable facts, clear entity, answer-first structure, consistency across the web |
| Role of links | Primary ranking signal | Secondary; mentions and consistent facts on third-party sites matter more |
| Role of markup | Rich snippets, nice to have | How the model learns who you are, what you sell, at what price |
| JavaScript rendering | Google renders it, slowly | Most AI crawlers do not: the bot sees an empty shell |
| How you measure | Positions, clicks, CTR | Citation rate: how many of N buyer questions name you across M assistants |
The last row is the one people skip. There is no Search Console for ChatGPT. You build the measurement yourself or you fly blind.
What does not change
GEO is not a reason to throw away SEO. Four things carry over untouched, and if they are broken, GEO will not save you.
- A site the bot can fetch: fast, server-rendered HTML, no login walls, no 403 for crawlers.
- One clear entity: the same name, role and offer on the site, in markup, on LinkedIn and on every directory.
- Content a human would actually use: specific, dated, with numbers. Models are trained to prefer it for the same reason people do.
- Patience measured in weeks. Google re-crawls in days; ChatGPT and Perplexity refresh their picture of you in 2–6 weeks after a fix.
If an agency sells you GEO as a separate universe with its own magic, ask what they will change on the page. If the answer is “content for AI”, keep your money.
What my own audits showed
In 2026 I offered free AI-visibility checks on Reddit and got 34 sites from Germany to Singapore. Six of the 34 blocked AI bots entirely with a hard 403 while Google got in fine, and the owners had no idea: it was a Cloudflare default. Seventeen of the 34 exposed no rating in their structured data although most had reviews. The record-setter, a renovation marketplace, served a bot 41 characters per page. Forty-one. Not one of these is a content problem.
My own site was the cobbler with no shoes. First deep GEO audit: 25 out of 100, and I build this for a living. Prerendered HTML, bots unblocked, full JSON-LD. Two weeks later: 64. No new articles, no links bought. Then a client catalog of 1,900 pages went from 58 to 79 in two months with zero lines of my code in their repository, only a 20-task spec their own developers implemented.
Meanwhile the economics shifted under everyone. On queries with an AI Overview, organic CTR fell from 1.76% to 0.61% over 2025, a 61% drop, while the visitor who does arrive from AI search converts about 4.4 times better. Fewer clicks, warmer clicks. That is the whole GEO vs SEO argument in two numbers.
Five decisions for the next quarter
Not a strategy. Five concrete calls, in the order I make them on a client site.
Open the door before you decorate the room
Check robots.txt, the firewall and Cloudflare bot rules for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. Fetch your own pages as those bots. If you get a 403 or an empty page, nothing below matters yet.
Serve HTML, not a promise of HTML
If the content assembles in the browser, prerender it or render it on the server. The bot does not wait for JavaScript. This one change took my own site from invisible to readable.
Write down who you are in machine-readable form
Person or Organization, Service with prices, Product with availability, FAQ that matches the visible text word for word. Markup is not decoration. It is the model’s source of truth about you.
Put one quotable fact on every page
A number with a date, a definition in one sentence, a comparison table. Models quote what is easy to quote. “Best-in-class solutions” is not quotable. “58 to 79 in two months on 1,900 pages” is.
Measure citation rate, not positions
Write 15–20 questions a real buyer asks. Ask them in ChatGPT, Perplexity and Google AI Mode. Count how often you are named. That number is your GEO baseline. Repeat monthly.
How to measure GEO without fooling yourself
Two rules I learned the hard way. First: brand queries do not count. Any site is found when you type its name. The test is the problem query: “who fixes X in Y”, “best tool for Z”. By name you exist, by problem you may not.
Second: one answer is not a fact. Assistants vary between runs and between accounts. A client once showed me a single ChatGPT screenshot as proof that “nobody gets recommended in our niche”. I ran the same ten questions twice each. Three competitors were named consistently. He was named zero times out of twenty. The screenshot was true and useless.
So the unit is N out of M: ten questions, two runs each, three assistants, sixty answers. Named in 4 of 60 is a number you can move and re-measure. “ChatGPT doesn’t know us” is a feeling.
The client with the agency on retainer is still on page one of Google. He is also still absent from the answer his buyers read first. Both things are true at once, and that is the only real difference between SEO and GEO.
Ranking got you the click. Citation gets you the shortlist. The shortlist is where deals are won now.
Want to know whether AI names you?
€150 express check, no call: I ask ChatGPT, Perplexity and Gemini ten of your buyers’ questions, twice each. You get who is named instead of you, which facts about you are wrong, and five fixes in priority order. PDF in two days.
Sources
- Aggarwal et al., “GEO: Generative Engine Optimization”, Princeton / IIT Delhi / Allen Institute for AI, 2024
- Seer Interactive, CTR on queries with AI Overviews, 3,119 queries, September 2025
- Ahrefs, top-position CTR across 300,000 queries, December 2025
- Author’s audits: 34 live sites (2026), own site 25 → 64 (2025), client catalog 58 → 79 (2026)
