SEO and GEO: what changes when the answer replaces the list
SEO optimises to be found in a list. GEO optimises to be cited inside an answer. What has evidence, what does not, and what we will not promise.
SEO optimises for you to be found in a list. GEO optimises for you to be cited inside an answer. That is the difference, and the rest follows from it: when the user stops seeing ten blue links and starts seeing a paragraph with three sources beside it, the object of the work stops being position and becomes citation.
GEO stands for Generative Engine Optimisation, and the term comes from a 2024 academic paper (Aggarwal et al., KDD 2024). It applies to systems that synthesise an answer instead of returning links: Google’s AI Overviews and AI Mode, ChatGPT, Perplexity, Claude, Copilot.
This article follows the rules it argues for. If it did not, it would not be worth publishing.
The unit of work moved from page to passage
In SEO the unit is the page and the target is position. In GEO the unit is the extractable passage, and the target is being picked as a source.
The consequence is counterintuitive: a page can rank first and never be cited, because the answer it holds was not isolable in the text. A model that synthesises needs to find a block that answers on its own, without forcing anyone to read the four preceding paragraphs to know what the subject is. A well-ranked text built as a spiral, delivering its conclusion only at the end, is hard to cite even when it is correct.
Hence the practical rule that changes writing the most: answer before you contextualise. The answer in the first paragraph of the block, the context after it. It is the inverted pyramid of journalism, and it works here for the same reason it works there.
GEO does not replace SEO, it sits on top of it
The market often suggests otherwise, so let us settle it here. Generative engines use search indexes. ChatGPT and Copilot lean on Bing, AI Overviews on Google. If technical SEO fails, GEO fails upstream: a site that cannot be indexed cannot be cited, and no amount of citable writing fixes a robots.txt that blocks access or content that only exists once JavaScript has run.
Anyone selling you GEO as a replacement for SEO is selling you the roof without the foundation.
What has evidence
From the research that coined the term and from public measurements in 2026, these are the tactics with support behind them:
- Cite sources. Named links to the origin of every claim that is not yours.
- Give numbers. A quantified fact is citable; a generality is not.
- Date things. An explicit date wherever the subject is volatile. Models prefer dated sources.
- Quote experts. A sentence attributed to someone with a name is worth more than the same sentence with no owner.
- Answer before contextualising. The answer first, the framing next.
- Start from the question. A block that opens with the question the reader would ask is easier to match against what they typed.
- Keep facts consistent across surfaces. The same fact identical on the site, on the Google Business Profile, on LinkedIn and in directories. It is the principle behind local SEO’s NAP, extended to every fact about the entity: when sources contradict each other, the model loses confidence in all of them.
And one that has no evidence: repeating keywords. It does not work on generative engines, and on classic ones it is penalised.
The llms.txt case, with the numbers in plain sight
llms.txt is a text file listing a site’s content in a model-readable format. It is frequently sold as a visibility lever. That is not what the data shows.
| What is measured | Value | Source and date |
|---|---|---|
llms.txt files receiving zero requests |
97% | Ahrefs, 137,210 domains with traffic in May 2026 |
| Direct requests to the file, across more than 500 million AI bot visits over 90 days | 408 | Limy.ai |
| Domains in the sample publishing the file | 28% | Ahrefs, with their own caveat that their customer base is more technical than the open web |
Google’s position is on the record. Gary Illyes said on 23 July 2025, at Search Central Live APAC, that Google does not support the file and does not plan to, and the AI features documentation states it in writing: “You don’t need to create new machine readable files, AI text files, or markup to appear in these features” (Google, page last updated 10 December 2025). John Mueller publicly compared it to the old keywords meta tag, the one search engines learned to ignore precisely because whoever publishes it controls it.
Several publications quote that Google sentence with Markdown in place of markup. The word on the page is the latter. We were caught by that chain ourselves, and only noticed by going to the original.
None of this means the file is useless. It remains a legitimate agent-to-agent signal: Claude Code and Cursor read it when the user mentions a domain, and Anthropic recommends it for that purpose. We keep ours correct and current for that reason, which is not the reason it is usually sold on.
What we will not promise
We say this up front, not in the footer of a proposal.
Visibility in generative engines is not deterministic. It varies by session, by geography and by model version. Two identical questions on the same day can return answers with different sources. Three concrete refusals follow from that:
- We do not promise positions. There is no position to promise, because there is no list.
- We do not promise traffic. A large share of generative answers produces no click at all. Your visibility can rise while sessions fall, and you are better off knowing that before you see the report than after.
- We do not promise a short horizon. Expect three to six months of consistent work before there is a trend worth reading.
What we sell is work and measurement, because that is what we control. We measure with a fixed set of questions, written before measuring and reused, engine by engine, with dates. We report trend and variation, never a position. Every report declares the instability, because a measurement presented as a ranking is misleading even when the numbers are right.
Frequently asked questions
Do I need to buy GEO separately from SEO?
No. In this house GEO is handled where SEO already lives, not as a separate service with its own invoice. The tactics overlap substantially, and splitting them into two contracts serves the supplier, not the client.
My site ranks on Google but ChatGPT never cites it. Why?
The most common cause is that the answer is not isolable: the content exists, but spread across several paragraphs instead of sitting in a block that stands on its own. The second most common is technical, with content assembled by JavaScript and absent from the served HTML.
Should I block AI bots in robots.txt?
That is a business decision, not a technical one, and it should be made with the file in front of you. Name them explicitly, whether to allow or to block, so that permission is a decision rather than an omission. Blocking them removes you from the answers you could have been cited in.
How long until I see results?
Three to six months of consistent work, and what you see is a trend, not a position. Anyone promising shorter timelines and guaranteed numbers is promising what they do not control.
What Syntropic does
We handle GEO where we already handled SEO, with the same technical checklist and the same content checklist, and with one principle we do not negotiate: saying what we will not promise is part of the proposal.
In practice that means making sure your site is crawlable and that AI bots are named in the file rather than authorised by forgetfulness, that structured data describes the right entity, that content answers the question in the first paragraph, that claims carry a source and a date, and that your company’s facts are identical on the site, on the Google Business Profile and in the directories you appear in.
In a market where plenty of people renamed SEO as GEO and raised the price, our differentiator is dull to advertise and easy to verify: we show you the measurement and tell you what it does not prove.
If you want to know where your site stands on this, the diagnostic is free and comes with no strings attached. We look at what is holding back your visibility, in classic search and in AI answers, and tell you frankly what is worth changing. If the right answer is to leave it as it is, we will tell you that too.