1.6


Every major website platform is being rebuilt around code and AI agents. Four questions decide whether that’s your problem or someone else’s.
Articles in this series
I’ve written about why this matters elsewhere — The Citation Economy covers how attribution replaced ranking, and The New Rules of Digital Discoverability covers the search shift. If you want the reasoning, start there.
This is the practical version. An hour’s work, a list of what’s wrong, and six things to fix.
Bring a notepad.
Open two or three assistants — ChatGPT, Claude, Perplexity, Google’s AI results — and run these, in a fresh conversation each time so nothing carries over.
Write down, for each: what it said, what it got wrong, and where it got it from.
Two things usually surprise people. The first is how much of the answer comes from directories, review sites and third-party mentions rather than their own website. The second is question 2 — whether you appear at all when someone asks generically rather than by name.
That second one is the whole game. Being described accurately when someone already knows your name is table stakes. Being present when they don’t is the opportunity.
The fundamental check. In your browser, view the page source of your homepage — right-click, View Page Source. Not Inspect; View Source. That’s roughly what a crawler sees.
Can you find your actual headline and body copy in there?
If yes, you’re fine. If all you see is a shell of scripts with no readable text, your content is being assembled in the browser by JavaScript, and most AI crawlers won’t run it — they fetch raw HTML and move on. This is common with sites built by AI code-generation tools that default to client-side rendering, and it makes you invisible to the exact audience this article is about.
Managed platforms generally ship server-rendered HTML, so this usually isn’t your problem if you’re on one. If you’re on a custom stack, check it today.
Pick five facts about your business: legal name, trading name, address, phone number, and what you actually do.
Now check them across your website, Google Business Profile, LinkedIn, Companies House, any directories you’re listed in, and any professional bodies. Note every disagreement.
Assistants assemble answers from all of these. When sources conflict, you get either the wrong answer or a hedged one, and neither helps.
Look at your site and ask: if someone wanted to know your prices, your service area, your qualifications or your process, could they find a plain statement of it?
“Get in touch for a quote” is not an answer. An assistant can’t relay it, so it relays a competitor who published a number instead.
In order of return.
Boring, unglamorous, and the highest-value thing on this list. Pick the canonical version of every fact and make every source match. Budget an afternoon.
One page each for the things people actually ask: pricing, service area, process, credentials. Direct answers in the first paragraph, then the detail. Put the answer where it can be lifted out.
Organisation and LocalBusiness schema at minimum, plus FAQ schema where you have genuine questions. The important word is matches — schema that contradicts your visible page is worse than none, because it makes you look unreliable to the systems you’re trying to convince.
If your text isn’t in the HTML, nothing else on this list will help. Static generation or server-side rendering, and re-test.
Your own website is one source among many, and assistants weight independent mentions heavily. Professional bodies, industry press, genuine reviews, podcasts, a client’s case study. This is slow, it’s the least controllable item here, and it’s often what separates the businesses that appear in question 2 from the ones that don’t.
A plain text file at your domain root summarising what you do and pointing to your key pages. Adoption is inconsistent and I’d treat it as cheap insurance rather than a lever — an hour’s work, might help, won’t hurt.
Diarise the five questions quarterly. Same wording, fresh conversations, note what changed.
This takes ten minutes and it’s the only way you’ll know whether any of the above worked, or whether something has drifted. Answers change as sources change, and you won’t be told.
Modest, slow improvement, and one or two things that turn out to be badly wrong and are fixed in an afternoon.
The genuinely common outcome of this audit isn’t a clever optimisation — it’s discovering that an assistant has been confidently telling people your old address, a service you stopped offering, or a phone number that rings out. That’s worth an hour of anyone’s time to find.
TL;DR
AI assistants now answer questions about your business before anyone visits your site, and they assemble those answers from sources you don’t control. This is the audit that tells you what they’re currently saying, and the six fixes that change it.
Mark Thurman


Get Your Instant Estimate
Find out your project cost in under 60 seconds.