SEO and GEO, explained for business owners

What changed

For twenty years the job was to appear on a page of ten links and get the click. That job still exists. But your customers now type a growing share of their questions into ChatGPT, Perplexity, Gemini or Claude instead. Is my land suitable for cabins. How much does a food trailer really cost. Which clinic near me does this. Google answers many of the same questions itself, in an AI Overview above the links. The person gets one synthesised answer with three to five sources named beneath it. They rarely go further.

Two consequences follow. First, informational traffic to websites is falling even where rankings are stable, because the answer arrives before the click. Second, being one of those three to five named sources is now worth more than a position on a page nobody scrolls. Search engine optimisation, SEO, is the discipline of earning the position. Generative engine optimisation, GEO, is the discipline of earning the citation. They are not rivals. They draw on the same work.

What Google rewards now

Google has said the same thing for years and says it more literally now. Write for people. Show first-hand experience. Be an identifiable source. Run a fast site that is easy to crawl. Earn references from sites that earned their own trust. The list of what gets penalised has been just as stable. Pages made at scale with nothing new in them. Links bought or exchanged to move rankings. Content written for the crawler rather than the reader.

For a business that means publishing what you know and others do not, structuring it so machines can read it, keeping the site technically clean, and letting references come from being worth referencing. Everything else is either table stakes or a risk.

How an assistant builds an answer

When you ask an assistant a question with a factual or local component, it usually does not answer from memory alone. It runs a search against a live web index, often Bing's or its own. Then it reads a handful of pages and writes a summary that names the ones it leaned on. Different assistants weight things differently and the details change often, but the pattern holds across all of them:

• It has to find you. A site missing from the indexes an assistant searches does not exist to it. Many businesses have never checked whether Bing indexed them properly.
• It prefers concrete answers. A page that states a price range, a date, a number with a method behind it, gets quoted. A page that says "contact us for a quote" gives the model nothing to use.
• It prefers consistency. One business name, one address, one description, the same everywhere: on the site, in structured data, in directories, on maps. Conflicting details lower confidence.
• It prefers sources other sources trust. Being referenced by industry publications, directories and communities where the question is actually discussed puts you where the retrieval looks.
• It rewards structure. Clear headings that match the questions people ask, schema markup, a summary near the top, and a machine-readable guide to the site's sources.

None of it is a trick. A careful human researcher rewards the same five things, which is rather the point.

Why original data beats everything else

A model cites what a journalist cites. A number, a date, a source. Say you are first to publish the average installed cost of a hospitality cabin in Florida, third quarter 2026, based on these projects, calculated this way. Everyone who writes about that topic afterwards has to point at you, assistants included, because there is nothing else to point at. Almost nobody in your industry has acted on this.

A data lab inside a business's website is our name for the three kinds of asset that do this. None of them is an article.

• A calculator that answers a question your customers currently work out by hand. What will this cost, how long will it take, does my situation qualify. Tools earn links naturally and keep people on the page, and both signals count.
• An index tracked over time, covering prices, lead times, availability or demand. Collected from public sources, aggregated, dated, updated on a schedule. Every update is a fresh reason to be cited, and the history becomes something nobody can replicate retroactively.
• A methodology page stating where the data came from, how it was processed, when it was last refreshed, and what it does not cover. Without it the number is cited once. With it, the number is trusted and cited repeatedly.

Two things make this harder than it looks. Publish the data under an open licence that requires attribution, which formally invites the web to cite you with a link, and it does. Gather only from public sources, respect their terms, name them. A number obtained improperly discredits the whole lab the day someone notices.

We say this with some standing: UNIKOST runs a glamping lab and a project calculator, and MOBITRUCK runs a data lab on food-truck economics. Both are companies in our group, and both are on the cases page with what we can show.

Where links belong

Links are still how trust moves around the web, and assistants inherit that judgement through the indexes they search. What has changed is which links count. A reference from an industry publication that cited your data counts. So does a byline article by your founder in a trade outlet with disclosure, and a mention in a community where practitioners answer each other. All three count for Google and for the assistants.

Links bought from networks cost you now rather than counting. So do placements without disclosure, articles spun across dozens of thin sites, and forum posts from accounts pretending to be customers. Platforms remove them, search engines penalise them, assistants stop citing the sites involved. If someone offers you "GEO" that turns out to be link seeding, walk away.

How to measure any of this

Rankings were easy to check. Citations need a method:

1. Write down thirty to fifty questions your customers actually ask before buying. Not keywords. Questions, phrased the way a person types them.
2. Ask each question, every week, in each assistant that matters to your market and in Google with AI Overviews on.
3. Record whether you are named, whether there is a link, and what the assistant said about you.
4. Report the share of answers that cite you, by assistant, over time. That share is the scorecard, and it moves slowly.

The usual instruments still matter alongside it. Search Console for what Google shows. Referring domains for what the web is saying. For a data lab, how many times the number was quoted and by whom.

A checklist you can run this week

1. Search your business name in Bing. If the results are thin or wrong, ChatGPT is working from the same picture.
2. Check that your name, address and description match exactly on your site, your Google Business Profile, your social profiles and the top three directories for your industry.
3. Find one page on your site that states a price, a range or a timeline in plain numbers. If there is none, that is the first thing to publish.
4. Ask ChatGPT and Perplexity your ten most important customer questions. Note who gets cited. That is your competitor list for GEO, and it may not match your competitor list for anything else.
5. Look at the sites cited in those answers. Which of them could plausibly reference you if you published something worth referencing?
6. Ask yourself what number your industry argues about and nobody has measured. That is your first index.
7. Check that every article on your site names a real author and a date.
8. Add structured data for your organisation and your FAQ, and a machine-readable source guide for assistants.
9. Retire any content that exists only because "we needed a blog post that week".
10. Give it six months before you judge results. Shorter windows mostly show noise.

Why this is a subscription and not a project

Assistants change how they retrieve, which indexes they use and how they cite, on a cadence of weeks. Platforms change their rules. Your data starts ageing the day you publish it. A GEO setup done once is worth less every month, and a data lab nobody refreshes stops being cited as soon as a newer number appears. Most of the work is maintenance, so we price it monthly and publish exactly what the month covers.

Review status and limits

Source-led explainer, edited by a person before publication and checked against our own editorial gates. It contains no ranking guarantee, no citation-frequency promise and no claim about any specific client result. Figures describing our own group are labelled as such. Where an assistant's behaviour is described, it reflects what we observed at the date above and can change without notice.

Glossary

• SEO. Search engine optimisation: earning positions in search results.
• GEO. Generative engine optimisation: earning citations in assistant answers. Sometimes called AEO, answer engine optimisation.
• Citation. An assistant naming your page as a source for part of its answer.
• Share of citations. The percentage of a fixed set of questions where you are cited, tracked per assistant over time.
• Entity. Your business as machines understand it: name, address, description, identifiers, kept consistent everywhere.
• Structured data. Markup that tells machines what a page contains: an organisation, a FAQ, an offer with a price.
• llms.txt. A plain-text file at the root of a site that tells assistants what the site is and where its best sources are.
• Zero-click. A question answered on the results page or in an assistant, with no visit to any site.
• Primary source. The first publisher of a fact or figure, the one everyone else has to cite.