GEO · Generative engine optimization
How do you get your business recommended by ChatGPT?
You cannot buy a place in an AI answer. ChatGPT assembles one from sources it can crawl and corroborate: your own site, your Google Business Profile, directories, third-party pages. The job is to let OpenAI's crawlers in, publish your answer as plain HTML text, and make one consistent description of your business exist everywhere the engine already reads.
By Florian V. · Last updated 23 September 2026. Every number on this page names its source and its year in the sentence.
Definition
GEO is not SEO with a new name. It is not a different job either.
GEO stands for generative engine optimization. The term comes from a research paper, Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, “GEO: Generative Engine Optimization”, presented at ACM SIGKDD (KDD 2024) in Barcelona in August 2024. It stands on the same foundation as SEO: pages a bot can read, a clear description of who you are, content worth quoting. It optimizes for a different unit, the sentence an engine can lift, not the position it can rank.
| SEO | GEO | |
|---|---|---|
| What you optimize | A page, so it ranks above other pages for a query. | A passage, so an engine can lift it into an answer and attribute it to you. |
| What you measure | Position, impressions, clicks, conversions. | Whether you are named, in which answers, next to whom. Clicks are a weak signal here. |
| Who has to read you | Googlebot, which renders JavaScript. | GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, which, per Vercel's December 2024 measurement, do not. |
| Where the win shows up | A results page the buyer scrolls. | Inside a paragraph the buyer reads instead of scrolling. |
| What it pays in | Traffic. | Presence in the shortlist, earlier than traffic. See the Cloudflare ratio below. |
| What is identical | Crawlable HTML, a clear entity, correct facts in several consistent places, and content that answers a real question. Most of the work does not change. | |
Why it is worth your time
The answer is eating the click.
People stop clicking when a summary appears
The Pew Research Center, in a study published on 22 July 2025 that tracked what people actually did rather than what they said, found that users clicked a result link on 8% of visits where an AI summary was shown, against 15% where there was none, and clicked a source inside the summary itself on 1% of visits. Pew is an independent institute measuring behaviour, not a vendor surveying opinions. That is why this is the only click number on this page.
But citation is not a traffic tap
Cloudflare's June 2025 figures put the ratio of pages crawled to visits actually referred back at roughly 1,700 to 1 for OpenAI and tens of thousands to 1 for Anthropic. Anyone selling GEO as a replacement for SEO traffic is ignoring that. Being cited buys you presence in the decision, your name in the answer, ahead of the competitor who is not there. It does not buy you a traffic channel.
What works
Five things, in this order.
The order matters more than the list. The first item is a prerequisite: skip it and none of the rest can be read at all.
Serve your text as HTML, not as JavaScript
Vercel measured AI crawler traffic across its network and nextjs.org and
reported on 17 December 2024 that “none of the major AI crawlers
currently render JavaScript”, naming GPTBot, OAI-SearchBot,
ChatGPT-User, ClaudeBot and PerplexityBot. They do fetch your script files
(ChatGPT: 11.50% of its requests; Claude: 23.84%) and treat them as dead
text. A client-rendered site is literally an empty document to them. Test it
with curl -s https://yoursite.com | sed 's/<[^>]*>//g'
and read what comes back.
Let the right crawlers in, and prove it in the logs
OpenAI's bot documentation states plainly that “sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers.” Perplexity's documentation asks you to allow PerplexityBot in robots.txt to appear in its results. Anthropic documents that its bots honour robots.txt directives. Training and search are separate user agents, so this is a decision you make per crawler, not one switch. Then read your server logs: hits from GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot and Googlebot are the only honest before-and-after measurement in this field.
Answer first, in 40 to 60 words, then cite and quantify
This is the one recipe tested under controlled conditions and published in a peer-reviewed venue rather than an agency blog. Aggarwal et al., “GEO: Generative Engine Optimization”, KDD 2024, benchmarked 10,000 queries and report that their top methods, Cite Sources, Quotation Addition and Statistics Addition, achieved a relative improvement of 30 to 40% on the Position-Adjusted Word Count metric and 15 to 30% on the Subjective Impression metric. The same paper found keyword stuffing performed worse than the untouched original. Narrative introductions get skipped; answers get lifted. This page is built that way on purpose.
Exist on the handful of domains these engines actually quote
Citations are concentrated, not spread evenly. Profound's analysis of 680 million citations from August 2024 to June 2025 found Wikipedia accounted for 7.8% of all ChatGPT citations, and Reddit for 6.6% of Perplexity's. Semrush, studying 230,000 prompts between 14 July and 12 October 2025, found LinkedIn the single most-cited domain in Google's AI Mode. A brand-new site will not be quoted immediately; a signed, useful answer on a platform already quoted can be. Two cautions: Semrush sells SEO software, and the same Semrush data shows Reddit's share inside ChatGPT collapsing between August and September 2025, so never build acquisition on one platform.
Keep doing classic SEO
“SEO is dead” is an agency sales line. Ahrefs, studying 4 million URLs cited in AI Overviews across 863,000 SERPs in a March 2026 study, found roughly 37.9% of those citations came from the query's own organic top ten, down from roughly 76% in July 2025, which Ahrefs attributes to Gemini 3 and more aggressive query fan-out. The share fell. The link did not break, and the fanned-out queries are Google SERPs too.
Our own site failed steps 01 and 02 on the same day we wrote
this page. The homepage returned 439 bytes whose entire body was <div id="root"></div>, the only word in the whole document was the title, and
OpenAI's training crawler was being turned away with a 429, three attempts
out of three. That is a host-level bot filter, not a choice. We fixed ours
first. This is the report you get on yours.
What does not work
Three things you will be sold this year. Do not buy them.
This section is the difference between a useful page and an agency page. Each of these is currently being packaged and invoiced by someone.
01
llms.txt
No system reads it. Google's John Mueller wrote on Bluesky on 17 June 2025: “FWIW no AI system currently uses llms.txt”, adding that the consumer chatbots fetch your pages, not that file. Gary Illyes confirmed at a Search Central Live that Google does not support it and has no plan to. Posting one costs ten minutes and does no harm. Paying for it is paying for nothing.
02
Schema markup, sold as a citation lever
The only controlled test says no. Ahrefs tracked 1,885 pages that added JSON-LD, each matched against three control pages from different domains at the same citation level beforehand, and reported in May 2026: +2.4% in Google AI Mode, +2.2% in ChatGPT, neither statistically significant, and −4.6% in AI Overviews. The widely quoted “cited pages have three times more JSON-LD” is a correlation: schema marks a well-built site, it does not cause the citation. Add it for rich results and entity clarity. This page has it, and we are telling you it is not why.
03
“AI Overviews markup”
It does not exist, and Google says so in its own documentation: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary” and “You don't need to create new machine readable files, AI text files, or markup to appear in these features.” Related: Google-Extended is not a crawler and not a Gemini ranking switch, Google documents it as an opt-out control for training and grounding, with no separate user agent string.
One more, on format: the numbers that circulate everywhere, “guides with tables are cited 67% of the time”, “articles over 2,900 words get 59% more citations”, “+317% with multimodal”, have no published sample, period or method. They are copied from agency blog to agency blog. We will not put them on this page, and you should not accept them in a proposal.
Questions
Straight answers.
What does GEO stand for?
GEO stands for generative engine optimization. It is the work of getting your business named and cited inside an AI answer, ChatGPT, Perplexity, Google's AI Overviews, instead of ranking a blue link. The term comes from a 2024 research paper, Aggarwal et al., GEO: Generative Engine Optimization, presented at the KDD 2024 conference.
Do AI crawlers read JavaScript?
Not the major ones. Vercel measured crawler traffic across its network and nextjs.org and reported in December 2024 that none of the major AI crawlers currently render JavaScript, naming GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot and PerplexityBot. They download your script files and treat them as dead text. Googlebot does render, which is why AI Overviews are the exception.
Should I block GPTBot?
Training and search are separate decisions with separate user agents. OpenAI's bot documentation states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. So you can refuse GPTBot, the training crawler, and still be citable, but blocking the search crawler removes you from ChatGPT search entirely.
Do I need an llms.txt file?
No system reads it. Google's John Mueller wrote on Bluesky in June 2025 that no AI system currently uses llms.txt, and that the consumer chatbots fetch your pages rather than that file. Publishing one takes ten minutes and does no harm. It should never be sold as a service or promised as a lever.
Does schema markup get you cited by AI?
The only controlled test we know of says no. Ahrefs tracked 1,885 pages that added JSON-LD, each matched against three control pages at the same citation level beforehand, and reported in May 2026 a 2.4% change in Google AI Mode and 2.2% in ChatGPT, neither statistically significant, and a 4.6% decrease in AI Overviews. Add structured data for rich results and entity clarity, not as a citation lever.
Is there special markup for Google's AI Overviews?
No, and Google says so in its own documentation: there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary, and you do not need to create new machine readable files, AI text files, or markup to appear in these features. Anyone selling an AI Overviews markup package is selling nothing.
Is GEO a replacement for SEO traffic?
No. Cloudflare's June 2025 figures put the ratio of pages crawled to visits actually referred back at roughly 1,700 to 1 for OpenAI and tens of thousands to 1 for Anthropic. Treat AI citation as presence in the decision, not as a traffic channel. Ranking on Google still feeds AI Overviews, so classic SEO stays part of the job.
Sources
Every claim above, and where it comes from.
Primary research and vendor documentation
- Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande, “GEO: Generative Engine Optimization”, ACM SIGKDD (KDD 2024), Barcelona, 25 to 29 August 2024. arxiv.org/abs/2311.09735
- Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results”, 22 July 2025. pewresearch.org
- Vercel, “The rise of the AI crawler”, 17 December 2024. vercel.com/blog
- OpenAI, bots and crawlers documentation. developers.openai.com
- Perplexity, bots documentation. docs.perplexity.ai
- Google Search Central, AI features and your website. developers.google.com/search
Industry measurements, read with the caveat stated
- Ahrefs, structured data and AI citations, 1,885 pages with matched controls, May 2026. Ahrefs sells SEO software.
- Ahrefs, 4 million AI Overview citations across 863,000 SERPs, study dated 2 March 2026.
- Profound, “AI Platform Citation Patterns”, 680 million citations, August 2024 to June 2025. Profound sells an AI visibility product.
- Semrush, 230,000 prompts and over 100 million citations, 14 July to 12 October 2025. Semrush sells SEO software.
- Cloudflare, crawl-to-referral ratios, June 2025.
- John Mueller (Google), Bluesky, 17 June 2025, on llms.txt.
Where a source sells a product in the field it is measuring, we say so in the same sentence as the number. Our own measurement of bva.marketing on 23 September 2026 is reproducible with one curl command, and we will run the same command on your domain.
Free audit
Find out what the machines see on your site.
We fetch your pages as Googlebot, GPTBot, PerplexityBot and ClaudeBot, and send you the status code and readable word count each one gets, in writing, usually within two business days. No call, no pitch deck.