How to spot a GEO agency that is about to rip you off

The internet has filled up with people selling GEO services. Or AI SEO, or AEO, or whatever they are calling it this month. The pitch is always the same: pay us and ChatGPT will recommend your business.

I should declare an interest. I have spent fifteen years doing SEO, most of it in the discount code and affiliate world, which is about as competitive as search gets. If you can hold rankings in a space Google actively takes aim at, you learn quickly what is real and what is noise. And I believe within a fairly short time most searching will be done by or through AI. Which means I am effectively one of these agencies now. So take what follows with that in mind.

But I keep talking to business owners who have been sold something they cannot evaluate, by someone using words they have never heard, and they have no way of telling whether they are getting value or being taken for a ride. So here are the warning signs. If the outfit pitching you trips any of these, ask harder questions.

Red flag one: they sell GEO as something separate from SEO

This is the most common one, and it tells you a lot.

The only way an AI can recommend you is if it can find you through a search engine first. That is the mechanism. When someone asks ChatGPT or Gemini or Google’s AI who to hire, the AI goes and searches, reads what it finds, and builds an answer from that.

So if you are invisible in search, you are invisible to AI. They are not two separate problems with two separate solutions. They are the same problem viewed from a different angle.

Google says this outright in its own guidance on AI features: optimising for its generative AI experiences is optimising for Search, and it is still SEO.

There are real differences in how an AI searches compared to a person, and those differences matter. But anyone presenting GEO as a brand new discipline requiring a brand new invoice is either confused or hoping you are.

What to ask: “How does this work alongside my existing SEO?” If the answer treats them as unrelated, walk.

Red flag two: they are not tracking what AI actually cites

This is the big one for me. Arguably the whole game.

Here is what happens when someone asks an AI to recommend a business. It does not run one search. It performs what is called a fan-out: it quietly generates somewhere between three and a dozen related searches, runs them all, sifts through possibly a couple of hundred results, and reads maybe a dozen pages properly.

Then it builds its answer from those pages.

So the question that decides everything is: which pages is it actually reading? If your brand appears on more of those cited pages than your competitor does, you get recommended. If it does not, they do. That is genuinely the mechanism, stripped of jargon.

Which means if an agency cannot tell you which pages are being cited for the questions your customers ask, they are working blind. They are guessing, charging you for the guess, and hoping something sticks.

There are tools that track this properly now. Some are better than others, and the space is moving quickly. What matters is that somebody is looking.

What to ask: “Which pages does AI currently cite when someone asks for a business like mine, and how are you tracking that?” A good agency will have an answer. A bad one will change the subject.

Why generic rank tracking is not enough

Some tools will show you how often you appear for common queries. That is better than nothing, but it misses the point.

People do not type common queries into AI. They type long, specific, oddly-phrased questions. “Who can come out tonight for a leak coming through my kitchen ceiling in Poulton.” Nobody is tracking that as a keyword, but it is exactly the kind of thing that turns into a job.

You need to know what is being cited for the prompts that actually matter to your business, not for a tidy list of head terms.

Red flag three: they describe GEO as a technical job on your website

If the pitch is mostly about things they need to install, adjust or configure, be careful.

Just like traditional SEO, the overwhelming majority of this comes down to content. Content on your website, and content on the other pages that AI reads when deciding who to recommend. If the campaign is not centred on creating and improving that content, it is not really GEO.

Worth saying plainly, since it comes up constantly: Google does not penalise content simply for having been written with AI. Its published guidance on AI-generated content is clear that appropriate use of automation is not against the rules. What it targets is using automation to churn out pages primarily to manipulate rankings. Quality and purpose, not production method.

Google has been unusually direct about this too. Its guidance explicitly says you do not need llms.txt files, you do not need to chunk your content into fragments, you do not need special markup or AI-specific rewrites. A fair chunk of the technical GEO menu is being sold for work Google says is unnecessary.

There are genuinely interesting technical things you can do around AI. Making your site usable by AI agents, wiring up booking and quoting so an agent can complete a task, all of that is real and increasingly worth doing. But it is a different job. It will not make a large language model recommend you.

What to ask: “What proportion of this work is content, and what proportion is technical?” If technical dominates, you are paying for the wrong thing.

Red flag four: they skip the boring bit

Before anyone sells you anything clever, your own house should be in order.

Get the fundamentals right and scrutinise everything. Site speed, indexing, titles, internal linking, the Google Business Profile, the content actually answering what people ask. Not because that alone wins you AI citations, but because it lets you say with confidence: well, it is unlikely to be that.

That is the real value of doing the boring work first. It gives you a priority list. You work down it, ruling things out, until you find the smoking gun that has been quietly costing you rankings, or the ticking time bomb that goes off the next time Google ships a core update.

Skip that, and you are buying clever tactics on top of broken foundations. Which is a good way to spend money and learn nothing when it does not work.

And a word on the latest miracle tactic

Fifteen years in a space Google actively targets teaches you one thing above all others: every spammy technique that works gets hammered eventually.

Not might. Does. The only variables are when, and how wide the blast radius is when it happens.

Look at what happened to the “best of” listicles. A pile of so-called GEO experts jumped on those, because AI was citing them heavily and they were easy to manufacture. Then Google took aim, and a lot of businesses who had paid to be on those lists found their placements worthless or worse.

Google’s own guidance on generative AI content spells out the mechanism: churning out pages at scale without adding value for users falls foul of its scaled content abuse policy. That is exactly what a manufactured listicle farm is, however it is dressed up in the sales deck.

The bit people underestimate is the collateral damage. When Google swings at a tactic, it rarely lands cleanly. Legitimate sites get caught in the blast. Businesses that were only lightly involved end up explaining to their customers why they vanished.

So when I see the latest social post proclaiming some clever new trick, and it smells fishy, I leave it alone. Not because I am timid, but because I have watched this film enough times to know the ending.

What to ask: “Is this something that still works in two years, or something that works until Google notices?” If they cannot answer that without hedging, you have your answer.

The uncomfortable summary

Four questions will tell you almost everything:

How does this integrate with my SEO? It should be one strategy, not two.
Which pages are being cited, and how do you know? If they cannot answer, they are guessing.
Is this mostly content work? It should be.
Does this still work in two years? Or only until Google notices.

None of those require you to understand the technology. They just require the person selling to you to be straight about what they are doing.

And the honest bit about my own trade

I said at the top that I am one of these agencies now, and I meant it. So it would be a bit rich to write all this without applying it to myself.

Everything I do on this is content and search, treated as one job rather than two invoices. I write about how citations actually work publicly, for free, so you can judge whether I know what I am talking about before you pay me anything. And every price is published so you can see exactly what you would be buying.

That is not me being noble. It is that this corner of the industry is filling up with people selling smoke, and the fastest way to tell the difference is whether someone will explain their work in plain English and put a number next to it.

If they will not do that for you, they probably cannot.

Want an honest assessment of where you actually stand with AI search? Get your free audit, or email hello@inkon.co.uk. I read every message myself.