Over the past year I've watched a lot of SEO agencies change their name, their homepage and their LinkedIn headlines. SEO became GEO, AEO or "AI Search". That part is fine: the shift is real, and I use the same words myself.
What worries me is what's inside the new offer. In most proposals I'm asked to review, "GEO" means content production: a fixed number of blog articles a month, topic clusters, glossaries, "ultimate guides". The promise has changed from "you'll rank" to "AI will cite you". The work hasn't changed at all.
A new sign, the same offer
Blog production is easy to sell. It's measurable in units, it's easy to staff with juniors, and it produces something to show every month. It also maps neatly onto a new metric: the number of times an AI answer links to your site.
The trouble is that for a company that sells products or services, that metric is a long way from money. A health brand doesn't earn anything when ChatGPT links to its article about what a fungal infection is. It earns when someone asks which treatment to buy and the answer names its product.
The goal of SEO was never traffic, and the goal of GEO isn't visibility. The goal is profit. Visibility only matters when it leads to a sale.
Cited is not the same as recommended
A brand can appear in an AI answer in three different ways, and they are worth very different amounts to a seller.
Three ways to appear in an AI answer
| What happens | Value for a seller | |
|---|---|---|
| Cited | Your page is linked as a source of information, often on a general question | Low to medium. Builds trust, rarely sells on its own |
| Mentioned | Your brand is named in a list or in passing | Medium. You're in the consideration set |
| Recommended | Your brand is the answer, with reasons: who it's for, why it's better | High. This is the moment of choice |
A recommendation doesn't need your page as the source. The model takes the reasons from wherever they are stated most clearly.
Citations are counted because they're easy to count: there's a link, a tool can see it. Recommendations are harder to track, because they live in the wording of the answer. That's exactly why so many offers optimise for the wrong one.
A brand AI recommends but rarely cites
Here's a pattern I saw recently in data for a consumer health brand present in seven European markets. In five of those markets, AI answers named the brand in roughly 40–50% of the tracked questions. Its own website was linked in only 17–20% of them.
In one market, the brand was the clear leader among the brands AI recommended. Among the websites AI cited as sources, it was second, behind the national public health service.
Same brand, two metrics
Named in answers — about 40–50% of questions in five of seven markets. In one market, first among recommended brands.
Website cited as a source — about 17–20% of questions. In one market, second behind the public health service.
Anonymised data from AI-visibility tracking across seven markets.
A content agency would read this as a problem: "your site is under-cited, let's write fifty medical explainers". I read it differently. The brand already wins where the money is. Trying to out-write a national health service on "what is it and what causes it" is expensive and mostly pointless. AI will keep citing the public authority on medical definitions, and it should.
The useful work is elsewhere: making sure the product pages state the reasons a model uses when it recommends (who the product is for, how it differs, how to use it), and that those reasons match what retailers and independent sources say.
What actually gets a brand recommended
When I trace why a model picked one brand over another, the reasons come from three places. None of them is "more blog posts".
- Pages where people buy. Product, offer and category pages that say in plain text what the product is, who it's for, what it does better and what it costs. These are usually the least AI-ready pages on a site, and the ones closest to revenue.
- Answers to buying questions. Comparisons, "which one is right for me", alternatives, pricing, fit and compatibility. Content that sits in the decision, not in the encyclopaedia.
- Confirmation from others. Retailers, reviews, media and expert sources that say the same thing about you. Models check whether others agree before they recommend, especially in health, finance and other trust-heavy categories. This is Digital PR and partner work: a team and a budget, not one person writing articles.
GEO work · serves sales or serves the report
Serves sales
- Product and offer pages rewritten with facts and reasons
- Comparison and "which one" content
- Prompts tracked by buying intent
- Recommendations measured separately from citations
- Same facts on your site, at retailers and in feeds
Serves the report
- A fixed number of articles a month
- Glossaries and definitions in a category you can't own
- Citation counts as the main KPI
- Informational prompts with no buyer behind them
- Money pages untouched after six months
When the blog does make sense
I'm not against blogs. I write one. But a blog for a company that sells has one job: to support the sale. In practice that means articles that sit close to a decision, answer the objections your sales team hears, and lead to an offer.
There are businesses where being cited is the product. Publishers, B2B consultancies, agencies, software companies selling to experts: for them, being the source AI quotes is part of how they win trust and clients. That's why I care about citations on this site. For a shampoo, a payment terminal or a hotel, it's a secondary metric.
And if the blog exists only because "AI needs content", it's worth asking what each article is supposed to sell. If nobody can answer, the budget is better spent on the pages where people already buy.
How to judge a GEO proposal
Before you sign, ask five questions. A good answer to each is short and specific.
- Which prompts will you track, and how many of them have a buyer behind them? "Best X for Y" and "X vs Y" matter more than "what is X".
- Do you measure recommendations separately from citations? If the report only counts links, it measures the easier half.
- What share of the work touches product, offer and category pages? If the answer is "we'll start with the blog", ask why.
- How will this connect to revenue? AI referral traffic in GA4, conversions from those visits, enquiries that mention AI.
- What will you stop doing? Good GEO work often means writing less, and fixing more.
If you already have a GEO proposal on the table, I'm happy to give it a second opinion. I'll tell you which parts move sales and which ones mostly move a dashboard. You can also see how I approach GEO and AI visibility and brand perception in AI.
Based on AI-visibility data and GEO proposals I have reviewed for brands in consumer health, FMCG and services; all examples are anonymised.
FAQ
Frequently asked questions
Being cited means an AI answer links to your page as a source of information. Being recommended means the answer names your brand as the choice, with reasons. A brand can be recommended without its website being cited, because the model takes the reasons from whichever source states them most clearly.
No. For most companies that sell products or services, the pages that matter most for AI recommendations are product, offer and category pages, plus comparison content. A blog helps when its articles sit close to a buying decision.
Usually not. On general and medical definitions, AI will keep citing public authorities, and competing with them is expensive. Focus on the questions where your product is the answer.
Track a set of buying-intent prompts and record whether your brand is named, how it is described and whether it is the recommended choice. Track citations as a separate, secondary metric, and connect both to AI referral traffic and conversions in GA4.
When expertise is what you sell: publishers, consultancies, agencies and B2B companies selling to experts. For them, being the source AI quotes builds trust and wins clients.
GEO builds on SEO, but the goal shifts from ranking a page to being chosen in an answer. The risk is buying the old content-volume model under a new name instead of work that changes how AI describes and recommends your offer.
