LLM Visibility
Audit
Your customers are asking ChatGPT, Perplexity and Google's AI Overviews about your category right now. We find out what those answers say, who gets named, who gets cited, and what it'd take to put you in the answer too.
~61%
CHATGPT'S SHARE OF AI CHAT TRAFFIC
~75%
AI OVERVIEW LINKS OUTSIDE TOP 10
4.4x
CONVERSION VS ORGANIC SEARCH
Sources: First Page Sage (May 2026 generative AI chatbot market share), Semrush (200,000-AI-Overview study), Semrush AI search traffic study (June 2025).
Your customer's first port of call is a chat box.
"Best CRM for a 12-person consultancy?" "Who does fractional CFO work in Manchester?" "What's the difference between X and Y?" These questions used to start in Google's search box. Now a lot of them start in ChatGPT or Perplexity, and the user sees three or four named options.
Either you're one of those names or you're not. There's no second page, no "click for more results", no scrolling. If the model doesn't pick you, the conversation moves on without you.
An LLM Visibility Audit tells you which conversations you're in, which you're losing, and what it'd take to be in the next answer too.
CLASSIC SEARCH
- 10 blue links per page
- Rank tracking on keywords
- Click-through driven by title and meta
- Position 4 still gets some clicks
ANSWER ENGINES
- One answer, three or four named brands
- Tracking on prompts, sentiment and citations
- The model decides who gets named, and why
- You're in the answer or you're invisible
Five things we measure, every time.
We don't ask ChatGPT about you once and write it up. The same prompts get run repeatedly across the engines your buyers actually use, so the numbers are repeatable instead of anecdotal.
Share of voice
For each prompt, who gets named? You vs the three or four competitors the model reaches for first. Counted, ranked, plotted over time.
Sentiment
When the model talks about you, what does it actually say? "Reliable, premium-priced". "Used to be good, declining". "Niche but trusted". We pull the language.
Citations
Which pages, Reddit threads, G2 reviews and news articles the model is reading to form the answer. Sometimes it's a site you've never heard of.
Hallucinations
Wrong pricing, discontinued products, made-up integrations, an ex-employee listed as your CEO. Catalogued by engine, with the exact wording.
Crawler access
Are GPTBot, ClaudeBot, PerplexityBot and Google-Extended allowed in? Is your content readable by them? Or have you quietly locked the door?
Where we come in.
Fixed price, no slide deck. You get a baseline you can rerun every quarter, a ranked list of fixes, and the work done on the ones you want done first.
We don't sell you a dashboard with a monthly fee. The audit is yours, the prompts are yours, the playbook is yours. You can run it yourself next quarter or ask us back.
BOOK AN AUDITPrompt set
We sit with your sales and marketing leads and build the prompt set: branded, competitor, category, problem-led, comparison. Usually 60 to 120 prompts. The ones a real buyer would ask, not the ones that flatter you.
Run and score
Each prompt runs ten times across ChatGPT, Google AI Overviews, Perplexity, Claude and Gemini. Answers get parsed for mentions, sentiment, citations and factual errors. You get a baseline you can rerun later and compare.
Diagnose the gap
We work out why. Wrong robots.txt rules, thin product pages, no schema, no third-party mentions, an out-of-date Wikipedia stub, a Reddit thread that's poisoning the well. Ranked by how much we'd expect each fix to move the score.
Make the fixes (optional)
If you want us to do the work, we will. Crawler access, schema, comparison pages, a proper /llms.txt, a Wikipedia edit you can defend, the press piece that gets cited. Priced per fix, no retainer required.
What you actually get back.
A single report you can hand to a CEO, a CMO and a developer and have all three understand their bit. Plus the raw data behind it.
The scorecard
One page. Your share of voice per engine. Sentiment score. Citation count. Top three competitors, ranked the same way. Plus the one number that matters most: how often a buyer leaves the conversation knowing your name.
The fix list
Ranked by expected impact. Quick wins (robots.txt, schema, a missing FAQ) at the top. Medium-lift content work in the middle. The slow, expensive bets (Wikipedia, earned press, Reddit credibility) clearly flagged at the bottom.
The raw data
Every prompt, every response, every model, every run. CSV and JSON. So you can argue with our scoring, run your own analysis, or hand it to your in-house team to rerun next quarter.
The rerun
A repeatable harness. Same prompts, same models, same scoring. Run it again in three months and you'll know whether the fixes worked, instead of guessing. You own it, you can run it without us.
Good. This isn't that.
SEO ranks pages against keywords. This measures whether a model picks your brand out of a category, what it says when it does, and which third-party sources convinced it.
Semrush studied 200,000 AI Overviews and found that only about a fifth to a quarter of the links cited overlap with the top 10 organic results. Ranking and being cited are not the same problem.
We won't tread on your SEO agency's toes. We'll hand them a list of pages the models can't read and let them get on with it.
SEO asks: "Where do we rank for ‘CRM for accountants’?"
An answer expressed as a position number on a results page.
We ask: "When ChatGPT recommends a CRM for accountants, are we on the list, and what does it say about us?"
An answer expressed in plain English, with sources and a sentiment.
Both matter. The audit tells you where they overlap and where they don't.
Some fixes help both. Some help only the AI side. We say which is which.
The ones we get asked first.
Does any of this actually move sales?
It moves consideration. People who land on your site from an AI chat tend to convert at a higher rate than cold organic traffic, because the model has already done the shortlisting. Whether that compounds into pipeline depends on how often you make the shortlist, which is what we measure.
Aren't the answers different every time?
Yes, that's why we run each prompt ten times and aggregate. One run tells you a story. Ten runs tell you a probability. The difference between "we're named in 7 out of 10 answers" and "we're named in 1 out of 10" is the difference between a known brand and a footnote, and you can't see that from a single chat.
Do I need an /llms.txt file?
It costs nothing to add one and it's becoming a polite convention. Be honest about what it does though: the proposal came from Jeremy Howard at Answer.AI in September 2024, and in June 2025 Google's John Mueller publicly said "FWIW no AI system currently uses llms.txt". So it's a future bet, not a today fix. Our audit will tell you whether yours is set up sensibly, and what the actual crawler logs say.
Which engines do you cover?
ChatGPT (with and without web search), Google AI Overviews, Perplexity, Claude and Gemini as standard. We can add Bing Copilot, Meta AI or You.com if your buyers actually use them, but for most UK B2B audiences those five cover it.
What if the audit finds the model is lying about us?
It probably will, in at least one place. Out-of-date pricing, the wrong founder, a feature you killed two years ago. We list each hallucination with the exact wording and the likely source. Fixing the source (your own pages, your Wikipedia stub, the Reddit thread) is usually how it gets out of the model the next time it's trained or it reads the web.
How much does it cost?
Fixed fee for a single brand and category. Scoped up front. Bigger if you've got several product lines or a long competitor list, smaller if you're a focused single-product company. The remediation work is priced separately, per fix, and you can do any, all or none of it.
Why you and not a GEO platform like Profound or Otterly?
If you want a dashboard you log into every Monday, buy one of those. They're good. We're for the company that wants a one-off audit, a ranked fix list, and the work done. Pick the dashboard if you want ongoing monitoring; pick us if you want answers and changes made.
How long until the fixes show up in the answers?
Crawler and schema fixes can show up in days for the engines that browse live (Perplexity, ChatGPT with search on, Google AI Overviews). Training-data changes take a model generation, which is months. We tell you which bucket each fix falls in so you don't expect overnight magic from the slow ones.
Find out what the answer engines say about you.
Tell us your brand, your category and the three or four competitors you care about. We come back with a sample run from one engine on a handful of prompts so you can see what it looks like. If it's useful, we scope the full audit.