What Happens When Your Customers Ask ChatGPT Instead of Google
Someone arrived on this site last month from a link inside a ChatGPT answer. One visit. That is not a trend and I am not going to pretend it is.
It was, though, the first time a machine had put my business in front of somebody who had not asked about me by name, and it sent me off to look properly at how these systems decide who to mention.
The short version: most of what gets you into AI answers is the work that already gets you found in ordinary search. There is no separate lever. There are a few things genuinely worth doing, and a growing amount of expensive nonsense being sold around them, so this post tries to separate the two.
What Is Actually Changing
Three things are happening at once and they tend to get muddled together.
Google now puts a generated summary above the results for a lot of queries. Assistants like ChatGPT, Claude and Perplexity will search the web mid-conversation and answer from what they find, with links. And people are starting to phrase things as questions to a person rather than keywords to a machine: who does small business websites near Shoreham and what would it cost, instead of web design shoreham.
What that means in practice is that for some queries the answer is assembled for the user rather than listed for them. If your business is not in the material the answer was assembled from, you are not in that conversation, even if you rank perfectly well on the page underneath it.
What Is Not Changing
More than the headlines suggest.
Your site still has to exist, be crawlable, and say clearly what you do and where you do it. Local intent still runs largely through maps and business listings. Reviews still count. Being genuinely well known in your area still counts, and arguably counts for more, because these systems lean on how consistently the rest of the internet describes you.
If your website is thin, out of date, or never states your prices or your service area, no amount of AI-specific tinkering will rescue it. That is the same advice as five years ago, which is either reassuring or boring depending on how much you were hoping for a shortcut.
How These Systems Decide Who to Mention
Broadly two ways, and it is worth knowing which one you are trying to influence.
The first is what the model absorbed during training. You have very little direct control over that and it is generally months out of date. Chasing it is not a strategy.
The second is retrieval: the assistant runs a search while it is answering, reads a handful of pages, and writes from those. This is the one worth caring about, because it works on the live web and it rewards pages that answer the question directly enough to be quoted.
A page that says "websites from £49 a month, covering Worthing, Shoreham and Brighton, usually live in three weeks" is easy to cite. A page that says "we craft bespoke digital experiences tailored to your unique vision" gives an assistant nothing to repeat, so it repeats somebody else.
Write the Answer, Not the Keyword
The practical change is smaller than it sounds. Where you used to think about the phrase somebody typed, think about the question they asked, and answer it in the first sentence.
- Use question-shaped headings where the question is a real one. "How much does a small business website cost?" rather than "Website Pricing Information".
- Answer immediately underneath, in plain language, before the context and the caveats.
- Include the specifics: numbers, prices, timescales, the towns you cover, what is and is not included. Specifics get quoted. Vagueness gets skipped.
- Keep one fact in one place. If your pricing is stated three different ways across three pages, you have made yourself an unreliable source.
None of this is in tension with writing for people. It is mostly just writing more plainly, which was worth doing anyway.
Your Facts Have to Agree Across the Internet
An assistant asked for your phone number will find several versions of it and has to pick one. So will a person, but a person notices the conflict and you get a puzzled phone call. A machine just picks.
So the boring work matters more than it used to. Your name, address, phone number, opening hours and service area should be identical on your website, your Google Business Profile, and every directory you appear in. Where they disagree, assume the wrong one is the one that gets repeated.
That is the same discipline as keeping your business listing accurate, and it now pays twice.
Structured Data, llms.txt, and What Actually Helps
Two technical things come up a lot.
Structured data, sometimes called schema markup, is the older one. It is a machine-readable summary sitting invisibly in the page code: this is a business, here is its address, this is an article, these are the FAQs. It has been useful for search features for years and it is a reasonable bet for AI retrieval too, because it removes the guesswork. Worth having. Not magic.
llms.txt is a proposed convention for a plain text file summarising your site for language models, similar in spirit to robots.txt. I add one to the sites I build, and I want to be straight about the evidence: it is a proposal rather than a standard, adoption by the major assistants is unconfirmed, and nobody can currently show you it working. It costs almost nothing to include, which is why I include it.
The Local Angle
For a local business, most of this still routes through the same place it always did. When an assistant answers a question about who does something nearby, it is usually leaning on mapping and review data rather than reading fifty websites.
Which puts your Google Business Profile back at the centre of it: accurate categories, real photographs, current hours, and a decent run of reviews.
That is not a new answer. It is the old answer, now doing an extra job.
What I Would Not Spend Money On Yet
A few things I would push back on if they were pitched to me.
- AI SEO packages that will not explain what is actually being changed on your site. If the deliverable cannot be described in a sentence, there usually is not one.
- Publishing large volumes of machine-written articles in the hope of being found by machines. It produces pages that say nothing memorable, and search engines have been getting steadily better at spotting them.
- Anything sold on a guaranteed mention in a particular assistant's answers. Nobody can promise that. Answers vary between users and change with every model update.
- Rebuilding a website that is working, on the grounds that it is not AI-ready. If it loads, it is crawlable and it says what you do, these systems can already read it.
What Is Worth Doing Now
A short and fairly dull list:
- Give every service you sell a page that names it and says what it costs, or how the pricing works.
- Add a real FAQ section answering the questions you get on the phone, in the words customers use rather than your own.
- State your service area explicitly somewhere. Do not make anyone infer it from your address.
- Check your name, address and phone number match everywhere they appear.
- Keep your business listing current.
- Add structured data if your site does not have it.
- Watch your referral sources for visits from assistants, so you find out what is happening instead of guessing.
That last one is how I noticed any of this in the first place, and it is worth setting up regardless.
Where This Leaves You
Somewhere between the two stories you are being told. AI search is not about to make websites irrelevant, and it is not a fad you can wait out either.
The useful way to hold it is that the bar for being findable has moved slightly, in a direction that favours businesses who state plainly what they do, what it costs, and who they do it for. If you have been meaning to fix that anyway, here is a reason to get on with it.
