Seven things have to be true before an AI model will name you. The audit scores each one red, amber or green, and the plan you get back fixes them in this order, because an early one being broken makes the later ones impossible.
An AI crawler is the software ChatGPT, Perplexity and the rest send out to read websites, the same way Google has always sent one. Thirteen of them matter, and we check each by name. If they’re being turned away, nothing else on this list can help you, and the block is almost always a security setting switched on by default that nobody chose.
JavaScript is code that assembles parts of a page after it opens, inside the visitor’s browser rather than on the server. A person never notices. The AI crawlers don’t wait for it to run, so anything built that way is simply not there as far as they’re concerned. A lot of modern sites put their most important words in exactly that place.
Markup is a small block of facts written into the page’s code and never shown on screen: your name, where you are, what you sell, what it costs. It’s how a model knows you’re a particular business in a particular town at a particular price, rather than a phrase it once read somewhere. Most sites carry none at all.
A model answers a question by finding a passage that already answers it, and quoting that passage. Flowing prose under a heading like "Our approach" gives it nothing to take. A heading that asks the question a buyer would actually ask, answered directly underneath, gives it something to take. This is why sites that read beautifully often get mentioned nowhere.
Models trust what several independent sources say the same way: company registers, directories, professional bodies, review sites, LinkedIn, forums. Where those disagree about your name, your address or what you do, you get harder to name with any confidence. This is the slow one and usually the worst score. It also covers name collisions, where something better known already owns your name.
Four unglamorous things. Whether you have a working sitemap telling crawlers which pages exist. Whether a missing page returns a proper error instead of quietly showing the homepage. Whether duplicate versions of a page point at one original. And whether your content sits on your main domain rather than a separate one. None of it is visible to a visitor, and each one quietly weakens everything above it.
The test the other six exist to move. Twenty to thirty questions written for your sector and your area, in the words a real buyer would type, put to Perplexity, ChatGPT, Claude, Gemini and Google’s AI Overviews. We record the answer each one gave and the sources it cited, so you can see exactly what a buyer would have seen.
Three reasons, and none of them is that it’s thinner than it looks.
Twenty to thirty questions, not a hundred and fifty. Enough to see whether you’re named for what people actually buy. Not enough to map a category. If you need that, buy the £3,500 audit from someone who sells it.
Most of the gap between a £750 audit and this one isn’t analysis. It’s the discovery call, the scoping document and the account handling. We took all of it out.
This practice is new, so the case studies are currently worth more to us than the margin. That will stop being true and the price will go up when it does. It hasn’t yet.
A free check gives you a score out of something, generated automatically, rarely grounded in much, and unable to tell you why. Here a person reads your site and reads what the models actually said about you, works out which fault is causing which symptom, and puts the fixes in the order that will move things fastest. Layers one to four are checks no free scan runs, and the name-collision work in layer five isn’t something any tool surfaces at all.
One document, written by a person who has read your site, not a dashboard export with your logo on it.
One verdict per layer, with the reason. Not a score out of a hundred: a score can be gamed, a red can’t be misread.
The exact reply each AI model gave and what it cited.
Who the AI models name instead of you. Usually the more useful of the two lists.
Sequenced by leverage and by how long each fix takes to show. Perplexity moves within days, ChatGPT and Claude in one to three weeks, Google AI Overviews in four to eight.
No implementation, no ongoing monitoring, no retainer. This is a diagnosis. You get the plan and can hand it to whoever does your website. Nothing rolls into a subscription and there’s nothing to cancel.
Five working days from payment. Layers five and seven are done by hand rather than by a tool, so five days is honest rather than padded.
No. The checkout asks for your website and what you do, which is everything the audit needs, and then the report arrives. Most people buying something like this would rather not sit through a discovery call to find out a price that could have been written down, so we wrote it down.
It’s the work of getting a business named and quoted by AI models when someone asks them a question, rather than ranked in a list of blue links. It’s usually shortened to GEO. It overlaps with SEO without being the same job: a search engine ranks pages, an AI names businesses, and if the web doesn’t agree about who you are, you don’t get named.
No. An SEO audit asks whether Google will rank your pages. This asks whether an AI model will say your name out loud when a buyer asks it a question. A site can rank perfectly well and be absent from every answer, which is where most businesses currently are.
Not as part of this. The audit is a diagnosis and the price reflects that. You get a plan in the order it should be done, and you can give it to whoever builds your website.
No. Everything is done from the outside, from what an AI model can see. That’s the point: we look at you the way they do. All we need is your domain and what you do.
No. Parallax Thinking Ltd isn’t VAT registered, so £499 is £499.
Full refund any time before we start work, no reason needed. Work starts the next working morning after payment, so until then your money comes back in full.