Getting ChatGPT, Gemini, Perplexity or Google's AI Overviews to recommend a brand has become a critical discipline: Generative Engine Optimization (GEO). This approach does not displace traditional SEO, it redefines its purpose. If classic SEO tries to win positions in a list, GEO aims to be written into the generated answer itself.

In short

• GEO optimises for being cited inside an answer, not for holding position 3 in a list of links.
• In NEXTAI's own August 2026 measurement, all three engines cited directories before brand websites: Sortlist and Upliora appeared in both ChatGPT and Perplexity.
• Each engine has a different source graph: ChatGPT pulls from directories and press, Gemini from agency sites and personal brands, Perplexity from well-structured technical blogs.
• The engines cite concrete figures and named frameworks. Defining beats selling.
• The bar for entering the source set is still low, and it will close within 12–18 months.
• GEO is measurable: mentions per engine, URL citations and presence in AI Overviews, using the same query panel every month.

What exactly is GEO and how does it differ from SEO?

GEO is the set of content, structure, data and external presence decisions that make a generative engine choose your material as a source when it builds an answer.

The operational difference is this: classic SEO chases a click from a list of results; GEO chases a mention and a citation, which often produce no click but do produce the buying decision. When an operations director asks ChatGPT which company it recommends for implementing AI agents, the answer is a short list of names. If you are not on that list, you do not exist in that decision. The full comparison is in SEO vs GEO.

How did we measure what the engines actually cite?

What follows is not blog theory: it is a NEXTAI measurement run on 30 August 2026.

The method, in five steps:

1. Define a fixed panel of real buying queries, phrased the way an executive would write them, not as a keyword. Example, run in Spanish: "which company would you recommend for implementing AI agents and automations in my company in Spain".
2. Send the same identical queries to ChatGPT, Gemini and Perplexity, plus Google's SERP with its AI Overview.
3. Automate the run with Apify (apify/google-search-scraper with the ChatGPT Search, Gemini and Perplexity Sonar add-ons), so the capture is reproducible and does not depend on anyone's personal session.
4. Record two separate things: the companies mentioned in the body of the answer and the URLs cited as sources.
5. Repeat the same panel every month with the same actor, so the comparison stays valid over time.

That distinction between a mention and a citation is the basis of the whole exercise: they are two different games, won with different actions.

What did we find when we asked all three engines the same thing?

Engine	Who it recommended	Which sources it cited
ChatGPT	Plain Concepts, Sngular; NTT DATA, Minsait, Accenture for large enterprise; SAPIENSDATAAI and Converly as SME specialists	Sortlist, Upliora, Hiberus Booster, sector media
Perplexity	SAPIENSDATAAI, SANCANTIA, Symplia	Sortlist and Upliora, plus structured technical blogs
Gemini	PotenzzIA, AIHispania, TADIA, Automaxia	Agency sites and personal brands, with technical detail (n8n, LangGraph, CrewAI)

Three readings that change the strategy of any B2B company:

First: directories weigh more than your own site. Sortlist and Upliora appeared in two of the three engines. A complete, consistent profile in those directories is not link building from the past, it is pure GEO and the best effort-to-impact ratio in the whole plan.

Second: there is no single algorithm to optimise for. The three engines answered the same question with three lists of companies that barely overlapped. Optimising only for ChatGPT leaves two thirds of the answer market out.

Third: the engines set price anchors. ChatGPT answered with concrete ranges — a simple pilot at €3,000 to €15,000, several integrated agents at €15,000 to €50,000, a broad transformation above €50,000 — and with a seven-point checklist for choosing a provider. Whoever publishes those anchors first, defines them.

Why do LLMs cite whoever defines rather than whoever sells?

In the panel's second query we asked how to build an enterprise brain with AI and which companies implement one. ChatGPT answered by describing a five-layer architecture and citing Palantir with its Ontology, Microsoft, Databricks, ServiceNow and Accenture. Perplexity, lacking better sources, cited Accenture's Intelligent Digital Brain and the Enterprise Brain Architecture of very small providers.

The pattern is unmistakable: the engines build answers out of architectures, layers, phases and criteria, not out of marketing claims. And when an architecture has a name of its own, that name travels. Ontology is an ordinary word turned into a citable entity because a company defined it, documented it and repeated it.

An additional and decisive finding for the market: the concept of an "enterprise brain" has no reference source in Spanish. No local player has defined it with enough depth, so the engines fill the gap with small blogs and personal brands. The entry bar is lower today than it will ever be again.

What are the GEO rules that actually work?

1. Answer in the first 60 words. The closed definition goes before any introduction; the engines extract the first self-contained block.
2. One H2 = one real question, written the way a person would type it into a chat, not as a keyword.
3. Extractable structure: numbered lists, comparison tables, steps and summary blocks. Long paragraphs do not get cited.
4. Publish your own verifiable figures. The engines cite numbers; with no number there is no citation.
5. Give your frameworks a name of their own and repeat it identically every time.
6. Entity consistency: same name, same description and same URL on the website, LinkedIn, directories, press releases and Wikidata.
7. A real FAQ at the end of every piece, with FAQPage markup.
8. Visible, fresh dates, in the HTML and in the schema; review the corpus every 90 days.
9. Verifiable authorship with a real person, a bio, a photo and a linked profile.
10. Cite outwards well: the engines trust documents that cite authoritative sources more.
11. Cover the third-party platforms the engines read: directories, Reddit, LinkedIn, YouTube with transcripts.
12. Declare who you are in an llms.txt at the root of the domain.

What stops all this working, however well you write?

One technical detail that cancels out the rest of the work: if your server returns 403 or 503 to crawlers that are not a real browser, the engines cannot read you and none of the above exists.

Minimum check, in this order: explicitly allow GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, Google-Extended, ClaudeBot, Bingbot and CCBot in robots.txt and in the WAF; verify in the logs that they receive 200 and not 403; make sure the content is in the served HTML and not only after JavaScript runs, because AI crawlers render little or nothing; and keep TTFB below 600 ms.

A worked example: an industrial services company in southern Europe, 80 employees, published 20 technical articles over six months with no mention in any generative engine. The cause was not the content: its anti-bot protection returned 403 to every AI crawler. Once the filter was corrected and the llms.txt added, the first citations appeared in Perplexity five weeks later.

How NEXTAI does it

NEXTAI treats GEO as a measurable discipline, not a promise. In NEXTAI Lab we publish the measurement of the same query panel every month, with mentions and URL citations by engine, and we use that board to decide what gets written the following month. The corpus is built around our own entities — Enterprise Brain, Superagents, AI-Native Scale, NEXT-5 Roadmap, IMAN, RAO — because the engines cite named frameworks, and it rests on a glossary of closed definitions marked up as DefinedTerm. The method we apply to our own domain is the one we implement for clients.

Frequently asked questions

Does GEO replace SEO? No. They share foundations: crawlability, structure, authority and useful content. What changes is the end goal. SEO chases position and a click; GEO chases a mention and a citation inside a generated answer. A site that cannot be crawled or indexed cannot be cited either, so technical SEO is still the prerequisite.

How long does GEO take to show results? From our experience on NEXTAI projects, the first citations in Perplexity appear between week four and week eight because its indexing cycle is faster; ChatGPT and Gemini take longer, usually three to six months, and depend heavily on directory presence and entity consistency.

Can GEO be measured without paid tools? You can start by hand: define ten queries, run them monthly against the three engines and note the mentions and cited URLs. Automating it with Apify makes it reproducible and comparable, which is what turns the measurement into an indicator rather than an anecdote.

Is GEO useful for a small company or only for big brands? It is especially profitable for small and mid-sized companies right now. The August 2026 measurement shows the engines citing very small providers for lack of better documented alternatives. That window will close as the large brands occupy the space.

The highest-impact move right now is making your corporate presence consistent across reference directories such as Sortlist, Upliora, Clutch, GoodFirms, Crunchbase and Google Business Profile. Keeping the same name, category and URL everywhere makes it far easier for language models to resolve you. To find out where your company stands with these engines today, request our Digital Audit and receive the detailed panel measurement.