Your Customers Are Asking ChatGPT About You. Is It Answering?
Your Customers Are Asking ChatGPT About You. Is It Answering?
By Aaron Kennedy, Founder of Originate. Last reviewed: 16 June 2026.
AI search optimisation is the work of making the answer engines, ChatGPT, Perplexity, Claude and the rest, know your business exists, hold your facts correctly, and put your name forward when someone asks about your category. The honest starting position for most UK businesses is that none of that is happening, because 87% of UK companies failed to appear in AI-generated recommendations for their primary commercial terms (per LLM Listed’s study of 500 UK business websites, June 2026), and below I’ll show you how to check your own score in ten minutes and what actually moves it.
On this page
- Your customers have already moved
- What AI discoverability actually means
- How invisible most businesses are
- The 10-minute self-check
- We ran it on ourselves and scored 24 out of 100
- What fixing it looks like
- What this costs, and who should bother
- Common questions
Your customers have already moved
Try this tonight. Open ChatGPT, type “best [whatever you do] near me”, and read the names that come back, because if yours isn’t one of them you’ve just watched a sale happen in a room you weren’t allowed into. And this is happening more often than the old habit of typing a query into Google and scrolling, which is the shift the numbers have started to catch. In a survey fielded this spring, 51% of B2B software buyers said they now begin their research with an AI chatbot more often than with Google, up from 29% a year earlier (per G2’s Answer Economy report, April 2026). That is a different front door to your category, and the founder who notices first tends to be the one who asked the machine about their own business and got an answer that never mentioned them. The behaviour moved before the search-volume tools caught up, which is why most agencies aren’t talking about this yet, and why your competitors mostly aren’t either.
What AI discoverability actually means
A small industry is growing up around this, and a lot of it reads like a digital tarot reading: vague promises, a fee, and no way to check whether anything happened. So here is the plain version. AI discoverability comes down to three things working together, whether the engines know your business is real, whether the facts they hold about you are correct, and whether they put you forward when a customer asks. The acronyms you’ll see thrown around all describe that same job, and generative engine optimisation, answer engine optimisation, AI SEO and AI search optimisation are different names for it. Strip the jargon and you’re left with the question Ben Harper of LLM Listed put best: “The question is ‘Does AI know we exist, trust us, and recommend us?’” Nothing mystical sits underneath that, because it comes down to facts a machine can read and content worth quoting, both of which you can check yourself before you pay anyone a penny.
How invisible most businesses are
The scale of the gap is the part that stops founders mid-sentence when I tell them at events. Across a study of 500 UK business websites in legal, financial services, software, healthcare, professional services and ecommerce, 87% of UK companies failed to appear in AI-generated recommendations for their primary commercial search terms (per LLM Listed’s study of 500 UK business websites, June 2026). Weigh that honestly, because LLM Listed sells AI discoverability services, so the figure comes from a vendor with a reason to alarm you, and I’d rather you knew that before you let it land. It lands anyway. Nearly nine in ten UK businesses, tested on the exact terms their own customers would type, simply weren’t in the answer. A separate dataset tells the same story from the other end, where the overlap between Google’s top-ranking pages and the sources AI answers actually cite collapsed from roughly 70% in early 2024 to under 20% (per 5W research with Brandlight, May 2026). Ranking on Google no longer carries you into the answer, so it’s a different contest now, and almost nobody has entered it.
The 10-minute self-check
You don’t need a tool, a login, or a consultant to find out where you stand, just ten minutes, a browser, and a willingness to read an answer that might sting. The check has three parts that build on each other: ask the engines what they know about you, confirm that the facts they’re reading agree with each other, and confirm that your website is even built for a machine to read. Run all three before you spend anything, because the result tells you whether you’ve got a real problem worth paying to fix or a site that’s already doing fine without you realising.
Ask the engines what they know about you
Open ChatGPT and Perplexity in two tabs and run three prompts in each. Start with the branded ask, “What can you tell me about [your business name]?”, where a good answer names you correctly, gets your services and location right, and ideally cites your own website, while a bad one confuses you with another company or invents details. Then run the category ask, “Who are the best [what you do] in [your town]?”, because that is the list you want your name to appear in. Last comes the problem ask, “How do I [the problem you solve]?”, where a strong result quotes a page from your site as part of the explanation. Read all six answers the way a stranger deciding whether to call you would, rather than as the owner who already knows the truth.
Check the facts the machines are reading
The engines build their picture of you from whatever they can find, so the real question is whether your name, services, prices and location say the same thing everywhere they look. Open your website, your Google Business Profile, and any directory you’re listed in, side by side, and compare them line by line, because if the address differs in three places, or your services list changed two years ago on the site but never on the profile, the machines see three slightly different businesses and end up trusting none of them fully. This part is thankless and it’s most of the battle. A model that can’t reconcile your facts tends to leave you out of the answer rather than guess.
Check whether your site is machine-readable
This part sounds technical and isn’t. Right-click your homepage, choose “View Page Source”, then search the code with Cmd+F or Ctrl+F for the word “schema” or the phrase “application/ld+json”, because finding a block of it means your site is labelling its facts for software to read, and finding nothing means the machines are guessing from raw text. Next, type your domain followed by /llms.txt into the address bar and see whether a plain page loads or you get a 404. Then glance at the dates on your key pages, since a page last touched in 2021 reads to an engine as stale. None of this needs a developer to inspect, even if fixing it later might, and that is a decision you make once you know the score.
Now do the easy version: run the Diagnostic on your site. It scores you across the five pillars and hands you the real fixes, free, with no call required.
We ran it on ourselves and scored 24 out of 100
I sell this work, so before I tell you to test your own site, here’s ours, measured on 10 June 2026 against the same five prompts, scored out of 100, and published with the bad parts left in. We came back at 24 out of 100, up from a measured 4 out of 100 in March, which is real movement in under ninety days and still nowhere near where a business that charges for this stuff ought to sit. The per-prompt breakdown is where the honesty earns its keep:
| Prompt | What we asked | Perplexity | Claude | Result |
|---|---|---|---|---|
| 1. Branded | ”Tell me about Originate” | 5/5 | 2/25 | Perplexity describes us accurately, with the right pricing and contact details, and cites weareoriginate.com. Claude’s training knowledge still confuses the name. |
| 2. Category | ”Best web design agency in Weston-super-Mare” | 5/5 | 0 | Perplexity lists us first and cites our site. |
| 3. Problem | ”How do I improve my website’s conversion?“ | 0 | 0 | No Originate content existed to cite, so generic sources won it. |
| 4. Comparison | Us versus a named competitor | 0 | 0 | Perplexity found nothing usable on either of us. |
| 5. Service | ”Pay monthly web design UK” | 0 | 0 | The query that fits us best, owned by a rival at £145 a month, with us absent. |
On the two prompts where a live engine reads our actual site we win cleanly, and on the three where it needs a page to quote from we score zero, because until recently we hadn’t written those pages. That gap is the point of this essay. We’re fixing ours in public against the same 5 Pillar framework the Diagnostic uses, and for the city context we publish the same dataset that sits behind the leaderboard: 9,565 Bristol sites scored. Average: 42/100. Only 6% of those pass Google’s speed benchmarks and the typical one takes 10 seconds to load on a phone, and nobody else in this category publishes their own score, which tells you something about how the category usually sells.
What fixing it looks like
The fixes are unglamorous, repeatable, and mostly free of magic, and here are the four that move the score, in the order they tend to matter.
One set of facts, everywhere
Decide the canonical version of your business, the exact name, the one-line description, the address, the services and the prices, then make every surface a machine reads match it word for word, from the website to the Google Business Profile to the directories to the social bios. When those facts agree, the engines stop hedging and start recommending, because confidence is half of what they’re quietly scoring. It’s the dullest item on this list, and it’s the one that moves the needle furthest.
A file that talks to the machines
An llms.txt file is a plain-text page at yourdomain.com/llms.txt that tells AI systems, in writing, what your business is, what it offers, and where the important pages live. Writing a good one takes about thirty minutes and costs nothing to host. Ours sits live at weareoriginate.com/llms.txt. It won’t single-handedly rescue a score, and it’s a clean, cheap signal that you’ve thought about being read by software, which is more than most of your competitors have bothered to do.
Markup the engines can actually parse
Structured data, also called schema markup, is a way of labelling the facts on your pages so software reads them without guessing: this is the price, this is the address, this is the review, this is the author. A page with it is one a machine can quote with confidence, and a page without it is a wall of text the engine has to interpret and often gets wrong, which is the same reason we build without a CMS and keep the markup clean underneath. The labelling is invisible to your visitors and decisive to the machines.
Content worth citing
AI answers quote specific, dated, attributed, data-heavy pages, the ones with a real author, a clear date, and numbers that trace back to a source. Generic agency copy that could describe any firm in any town never gets quoted. There’s nothing in it worth lifting. This essay is built to that standard on purpose, named author, visible date, every external figure sourced, which is also why our own problem-query score sat at zero until we started writing pages like this one.
What this costs, and who should bother
Plainly, not every business needs to pay for this yet, and saying so probably sells more of the ones that do. If you’ve no content and you’re still finding your feet, this isn’t your priority, so get a fast, well-built site live first and come back to discoverability once you have something worth recommending. For the business that already has customers asking the engines about its category, the maths is straightforward. Standalone SEO retainers in the UK start around £500 a month, with meaningful strategic work running £800 to £1,500 a month (per Whitehat SEO and Red Eagle Tech pricing guides, 2026), usually with AI discoverability bolted on as an extra or not offered at all. We bundle it into Foundations Launch & Perform, £247 a month including VAT, which includes a quarterly AI discoverability Diagnostic, a monthly keyword brief you can read in two minutes, and one new content page a month built from your own data. That last one is the citable kind that closes the problem-query gap I just showed you on our own scorecard. The score that tells you whether you need any of it is free, so start there.
Run the Diagnostic on your site. You get a real score across the five pillars and the actual fixes, free, with no call required, because the score does the persuading and we don’t need to.
Common questions
How do I check if my business shows up in ChatGPT?
Open ChatGPT and Perplexity, then run three prompts in each: the branded ask (“Tell me about [your business]”), the category ask (“Best [what you do] in [your town]”), and the problem ask (“How do I [the problem you solve]?”). If your name is missing or the details come back wrong, you’ve found your gap. The full ten-minute method, including the facts check and the machine-readability check, is in the self-check above, and the easiest version is to let the Diagnostic run all of it for you.
How do I get my business recommended by ChatGPT?
There are no tricks here, and anyone selling you one is selling snake oil. You make your facts identical everywhere a machine reads them, add structured data so software can parse those facts cleanly, publish a plain llms.txt file, and write content specific enough to be worth quoting. The engines put forward businesses they can read clearly and lift a sentence from, so the whole job is giving them that rather than hoping it happens by accident.
Is my business visible in AI search?
The honest answer is that you can’t know until you check, and most founders assume they’re fine because they rank on Google, which no longer follows. The fastest read is the branded ask in ChatGPT and Perplexity, where a correct, cited answer means the engines have you, and a vague or wrong one means they don’t, and the Diagnostic turns that hunch into a score across the five pillars so you’re not guessing from a single prompt.
Is AI discoverability the same as SEO?
It grew out of SEO and it’s no longer the same job. Ranking on Google used to mean appearing in the answer, and now the overlap between Google’s top results and the sources AI answers cite has fallen to under 20% (per 5W research with Brandlight, May 2026). Classic SEO still matters for the people who type into a search box, while a separate set of moves now decides whether you turn up at all when someone asks a machine instead.
What is llms.txt and does my business need one?
It’s a plain-text file at yourdomain.com/llms.txt that tells AI systems what your business is, what it offers, and which pages matter, a sort of machine-readable equivalent of a short, honest “about us”. Writing a good one takes about thirty minutes, and most businesses don’t have one yet, so it’s a cheap, early signal that you’ve thought about being read by software, worth doing even though it won’t carry your whole score on its own.
Does ranking on Google mean AI tools will mention my business?
No, and this is the assumption that catches the most founders out. A page can sit at the top of Google and never get cited inside an AI answer, because the engines pull from sources a Google ranking no longer predicts well, and they’re separate contests with separate rules. A strong Google position is genuinely useful, and it isn’t a substitute for being something the answer engines can read and recommend.
How long does it take to show up in AI answers?
Honestly it depends on which engine, and I won’t promise you a date. Perplexity reads live pages, so a clean, well-marked-up page can start showing in answers within weeks, while training-knowledge models lag a long way behind because they update on their own schedule, with Claude’s built-in memory sitting months back. The only timeline I’ll stand behind is our own measured movement, 4 out of 100 to 24 out of 100 in under ninety days, with the live-engine prompts moving first.
Run the Diagnostic on your site. Find your score across the five pillars, see the real fixes, and decide for yourself, because it’s free and the 24 out of 100 we just published is the only sales pitch we’ve got.
Written by Aaron Kennedy, Founder of Originate. Last reviewed 16 June 2026. Every external figure on this page links to its source, while our own score (24/100, measured 10 June 2026) and the Bristol dataset are first-party and reproducible. This page sits in our monthly five-prompt test rotation and will be re-dated when the next baseline lands.
See also: why we build without a CMS, Core Web Vitals and the Bristol reality, and every essay in /writing.
Common questions
How do I check if my business shows up in ChatGPT?
Open ChatGPT and Perplexity, then run three prompts in each: the branded ask ("Tell me about [your business]"), the category ask ("Best [what you do] in [your town]"), and the problem ask ("How do I [the problem you solve]?"). If your name is missing or the details come back wrong, you've found your gap. The full ten-minute method, including the facts check and the machine-readability check, is in the self-check above, and the easiest version is to let the Diagnostic run all of it for you.
How do I get my business recommended by ChatGPT?
There are no tricks here, and anyone selling you one is selling snake oil. You make your facts identical everywhere a machine reads them, add structured data so software can parse those facts cleanly, publish a plain llms.txt file, and write content specific enough to be worth quoting. The engines put forward businesses they can read clearly and lift a sentence from, so the whole job is giving them that rather than hoping it happens by accident.
Is my business visible in AI search?
The honest answer is that you can't know until you check, and most founders assume they're fine because they rank on Google, which no longer follows. The fastest read is the branded ask in ChatGPT and Perplexity, where a correct, cited answer means the engines have you, and a vague or wrong one means they don't, and the Diagnostic turns that hunch into a score across the five pillars so you're not guessing from a single prompt.
Is AI discoverability the same as SEO?
It grew out of SEO and it's no longer the same job. Ranking on Google used to mean appearing in the answer, and now the overlap between Google's top results and the sources AI answers cite has fallen to under 20% (per 5W research with Brandlight, May 2026). Classic SEO still matters for the people who type into a search box, while a separate set of moves now decides whether you turn up at all when someone asks a machine instead.
What is llms.txt and does my business need one?
It's a plain-text file at yourdomain.com/llms.txt that tells AI systems what your business is, what it offers, and which pages matter, a sort of machine-readable equivalent of a short, honest "about us". Writing a good one takes about thirty minutes, and most businesses don't have one yet, so it's a cheap, early signal that you've thought about being read by software, worth doing even though it won't carry your whole score on its own.
Does ranking on Google mean AI tools will mention my business?
No, and this is the assumption that catches the most founders out. A page can sit at the top of Google and never get cited inside an AI answer, because the engines pull from sources a Google ranking no longer predicts well, and they're separate contests with separate rules. A strong Google position is genuinely useful, and it isn't a substitute for being something the answer engines can read and recommend.
How long does it take to show up in AI answers?
Honestly it depends on which engine, and I won't promise you a date. Perplexity reads live pages, so a clean, well-marked-up page can start showing in answers within weeks, while training-knowledge models lag a long way behind because they update on their own schedule, with Claude's built-in memory sitting months back. The only timeline I'll stand behind is our own measured movement, 4 out of 100 to 24 out of 100 in under ninety days, with the live-engine prompts moving first.
