What ChatGPT Sent Our Agency Site: 8 Visits, 2 Enquiries, and What We Can't Prove
Why We Are Publishing Our Own Numbers
Most agency case studies are about a client, and the reader has to take the agency's word for the figures. This one is about our own site, measured in our own analytics, with the limits written next to the numbers. It replaces three anonymous case studies that we removed from this site because we could not back them with client data we are able to publish.
- 8 visits from ChatGPT out of 191 sessions in total (4.2%)
- 0 visits from Perplexity, Claude, Gemini or Copilot
- Landing pages: the homepage 5 times, About once, Pricing once, and once with no page recorded
- Where from: six visits from India, two from elsewhere
- Engaged time: three visits stayed two minutes or more (the longest eight minutes); four left within 15 seconds
- When: 3 visits between 18 and 21 September, 1 on 11 October, then 4 between 30 October and 3 November
The Two Enquiries, and How Far the Evidence Goes
On 2 November a founder submitted the contact form on our homepage. GA4 recorded that submission as a lead event inside a session whose source was chatgpt.com, which is direct evidence the visit came from ChatGPT. On 3 November the WhatsApp button, live since 17 October, recorded its first click, also inside a ChatGPT session. We do not know whether that was the same person.
The 18 September enquiry is weaker evidence. It arrived by email, so GA4 cannot connect it to a visit. It landed within the same hour as a ChatGPT visit to the homepage, which is suggestive and nothing more. When we asked the 2 November enquirer on 3 November how they found us, they told us they found us through ChatGPT, which turns that attribution from inference into something they said themselves. We have not yet asked the 18 September enquirer, so that one is still inference.
What We Changed, and When
No ChatGPT visit was recorded between 4 July and 17 September. The first four rows below came before the first visit; the last came after.
- 31 May: an llms.txt file, a dedicated About page for the founder, and the homepage content placed in the page's raw HTML so crawlers that do not run JavaScript can read it
- 4 July: robots.txt explicitly allowing AI crawlers, including GPTBot, OAI-SearchBot and ChatGPT-User
- 4 August: structured data for the organisation, the founder and the site linked into one graph, and descriptive headings
- 6 August: a machine-readable agent.json and a markdown version of every page, generated from the same data as the site, with a build check that fails if they disagree with the real prices and facts
- 17 October to 3 November: a WhatsApp contact with its own structured data, Bing crawl checks and resubmission of stale pages (ChatGPT search is widely reported to draw on Bing's index), and our LinkedIn company page tied into the site's structured data
What ChatGPT Says About Us When Asked
On 3 November we ran six buyer-style prompts in fresh ChatGPT chats with memory and personalisation switched off and web search on. We were named first for one of them: a startup founder who wants a marketing partner that owns growth, SEO and content, on a small monthly budget with an India-based team preferred. ChatGPT cited our published prices and our outcome-focused positioning. We were not named for the other three discovery prompts (best growth agencies for AI and SaaS startups in India, agencies that do AEO, and an agency for Instagram, Meta ads and SEO for an AI product in the US and UK). Competitors with named clients and third-party profiles were listed there instead.
Asked directly whether we are a credible agency, ChatGPT described our services and prices accurately and gave a preliminary 6 out of 10, "promising but not independently verified". Its reasons were fair: anonymous case studies it could not check, two different sets of Sanchara figures on different pages with no stated measurement windows, and no independent reviews or listings. Since then we have labelled the measurement windows, removed the anonymous case studies and linked our LinkedIn company page to the site. We will re-run the same prompts and see whether anything moved.
What This Does Not Prove
Eight visits is a very small number, and one more or one fewer changes the story. ChatGPT does not pass the question that led to a visit, so GA4 can see only that the visit came from chatgpt.com, never why. We cannot tie either enquiry to a particular change in the list above, and we cannot rule out that these people would have found us another way. The timeline is consistent with the changes helping, and equally consistent with changes on ChatGPT's side.
It is also one small site. A client site with different content, competition and audience will behave differently, and nothing here is a forecast for anyone else.
What We Will Do Next
We will ask the 18 September enquirer how they found us, so the second attribution stops being inference. We check ChatGPT referrals in GA4 every week and watch how often Bing recrawls our key pages. In December we will re-run the same six prompts and publish what changed, including if it got worse.
FAQ
Does ChatGPT send The Social Algorithm customers?
It has sent 8 visits since Google Analytics 4 began recording on 4 July 2026, and two enquiries coincided with ChatGPT visits. One enquirer has confirmed they found us through ChatGPT; the other has not been asked. That is too few to call it a channel.
What did you change to be readable to AI assistants?
An llms.txt file, explicit crawler access in robots.txt, page content present in the raw HTML, linked structured data, an agent.json and markdown version of each page generated from one data source, and fresher crawl data in Bing. The dated list is in the article.
Can you guarantee a client will be cited by ChatGPT?
No. Nobody can: retrieval systems change, and a citation today does not guarantee one tomorrow. We can make a site readable, consistent and measurable, and report what happens, including when it is nothing.
How do you measure AI search visibility?
Three ways: the GA4 session source (chatgpt.com, medium ai-assistant), fresh unpersonalised prompt tests with web search on, and Bing Webmaster crawl data for the pages those visitors land on.
Related reading
Sanchara Travels case study: 196 enquiries in 28 days · What a growth marketing agency costs in India · Algorithm-first marketing vs traditional agencies