What did we test?
On 19 October 2026 we asked six AI tools the same 37 questions a buyer might ask when looking for a web design, SEO or automation agency: ChatGPT, Claude, Perplexity, Gemini, Microsoft Copilot and Google AI Mode. Each question went into a fresh chat with memory switched off. That gave us 222 answers, and we logged every business each answer named and every source it used.
The questions fell into six groups, from asking about us by name to the problems buyers describe in their own words, such as “my website isn’t generating enquiries”.
Did the AI tools recommend us?
When the question used our name, almost always. When it did not, almost never.
| Question type | Questions | Answers that named us |
|---|---|---|
| Brand: “What is Proven Path Strategies?” | 1 to 5 | 28 of 30 |
| Local: “Web design agency in St Albans” | 6 to 13 | 7 of 48 |
| Sector: law firms, accountants, clinics | 14 to 20 | 0 of 42 |
| Problem: “My website isn’t generating enquiries” | 21 to 27 | 0 of 42 |
| Capability: GEO, n8n, Wix Studio | 28 to 32 | 1 of 30 |
| Competitor comparisons | 33 to 37 | 0 of 30 |
| Every question without our name in it | 6 to 37 | 8 of 192 (4%) |
The eight mentions came from Google AI Mode (3), Claude (2), Copilot (2) and ChatGPT (1). Perplexity and Gemini did not name us once. Copilot was personalised to our own account for most of the run, so we treat its results with caution.
The best single result came from Google AI Mode. Asked for agencies that build websites and automate business workflows, it named us first, and it took that from our Google Business Profile.
What did they get wrong about us?
Even when the tools named us, they often had old facts. Twenty two answers placed us in Harpenden, our previous address, which still appeared on our Business Profile and at Companies House. Nine described an old positioning we have since moved on from, which still lived on old social profiles, a directory listing and a domain we no longer use.
The lesson is simple: AI tools repeat whatever the web says about you, and the web remembers things you have long moved on from.
Which businesses got recommended instead, and why?
We did not name the agencies here, but the patterns were clear. Four things held across all six tools.
1. The same facts everywhere
Name, address, phone and a one line description that match on your website, Business Profile, Companies House, social profiles and directories. Every tool’s own how to answer said this, and our 22 Harpenden answers show the cost of getting it wrong.
2. A page that answers the exact question
This was the strongest pattern in the data. On “my website isn’t generating enquiries”, the tools cited small agencies purely because they had a blog post with that title. A solo freelancer was named on three questions through pages each written for one sector. If no page on your site matches the question, you will not be in the answer.
3. A complete Google Business Profile
Google AI Mode built its local answers from Business Profiles, showing ratings, review counts and addresses. It was behind every Google AI Mode mention of us.
4. Proof from other websites
The tools trust what others say about you more than what you say about yourself. Claude and Perplexity repeatedly flagged claims a business made only about itself, and most of the tools said they could find no independent reviews of us. Directories, review sites and “best of” lists did a lot of the work.
What else made a difference?
- Specific beats general. Pages aimed at one sector, platform, town or problem got cited. General “we do web design” pages almost never did.
- Published prices get quoted. Agencies that publish “from” prices were quoted with them. Agencies without prices got less detail.
- Each tool has its own habits. ChatGPT used the asker’s location for local questions. Google AI Mode read Business Profiles. Perplexity leaned on articles and directories. Claude was the most sceptical and checked Companies House. Gemini rarely showed its sources.
- Original data stands out. Most guides repeat the same unchecked statistics. A real before and after is rare, which is why we are publishing ours.
What are we doing about it?
This website is the first part of the answer. Following our own plan, we have:
- started correcting our address and description everywhere, using one sentence word for word;
- built a page for each service AI tools asked about, including AI search optimisation, n8n automation, AI chatbots and Wix Studio SEO;
- published our prices on a pricing page;
- written pages for accountants, law firms, financial advisers and private clinics, and for St Albans and the towns around it;
- begun work on independent reviews and directory listings.
Our targets for day 90: named in 30 or more of the 192 non brand answers, 10 or more of the problem and capability questions, no answers placing us in Harpenden, and at least five independent reviews.
How reliable is a test like this?
It is one run, on one day, from one location, with each question asked once. AI answers change between runs, so judge progress across the whole set of questions, never a single answer. We plan to ask the same 37 questions again, the same way, and publish the before and after here.
How can you check your own business?
Our free AI visibility check runs a smaller version of this test for your business: eight questions across four AI tools, with a report by email. The full audit asks 25 questions written for your business in four AI tools, three times each, with the marking checked by a person.