In the summer of 2026 we noticed an odd source in our analytics: people came to webnostra.pro from ChatGPT and went straight to Telegram. By October it was the site’s main organic channel. Below is what we learned in those months: what decides whether an AI assistant names a company, how to measure it and how long it takes. Every figure is from our own website, with a source and a date.
Why it matters for a business
Someone who asked ChatGPT “which agency does advertising for a developer in Phuket” arrives on the site with a recommendation already. They do not compare ten links; they check one. So these visits end in a contact more often than search visits.
Our figures for Oct 1-16, 2026 (Yandex Metrica, webnostra.pro, website segment):
| Source | Visits | Contacts (form, Telegram, WhatsApp) |
|---|---|---|
| ChatGPT and other AI assistants | 25 | 10 |
| Search engines (Google, Yandex, Bing) | 33 | 1 |
| Whole website | 377 | 22 |
In August and September AI brought 5 contacts a month; in October, 10 in the first 16 days. Thailand-related pages brought the most AI contacts: 15 visits and 7 contacts from Aug 16 to Oct 16. These are Russian-speaking business owners abroad looking for a contractor in their own language.
This is our website and our niche, and another business will see a different AI share. The pattern is clear, though: the channel is small in visits and large in leads.
How an AI assistant decides whom to name
Assistants with web search do not make the answer up: they search for pages, read them and assemble an answer with links. That gives three conditions.
- Your page has to be in the index of the search engine the assistant uses. ChatGPT and Copilot rely heavily on Bing, Gemini on Google; in Russia, Yandex’s Alice relies on Yandex.
- The page has to answer the question. Concretely: what you do, for whom, what it costs, what the results are and where to check them.
- Others have to talk about the company. Directories, industry media, reviews. An assistant finds it easier to name a company it sees in several independent sources.
What we did on our own website
The Bing index
Most companies look only at Google Search Console. For ChatGPT, Bing matters more. We connected Bing Webmaster Tools and found that as of Oct 16, 2026 Bing knew 171 pages of the site and kept 152 in the index, while pages published after Oct 6 got the status “URL cannot appear on Bing”. Resubmitting did not help: Bing waits for external links.
What works: the IndexNow protocol (it tells Bing and Yandex about new pages right after publishing; we send it on every site update), submitting pages in Search Console and Bing Webmaster Tools, and external links to new sections.
Facts with a source in place of promises
At our baseline measurement on Oct 11, 2026, Yandex’s assistant Alice, widely used in Russia, quoted us where the page had a price table and cases with figures.
After that we went through about 75 pages and articles and removed figures without a source: project counters without proof, ROI without a calculation method, market averages without a link. In their place we put figures from cases with a link to the case. An assistant quotes more readily what it can check, and a client trusts more readily what they can check themselves.
A short answer block visible to everyone
Below the first screen of every sales page and article there is a short answer: the page’s question, a paragraph with the main figure and 4-8 facts. It is the most quoted fragment of the page.
In October we discussed hiding this block from people and showing it only to AI. We decided against it: different text for people and robots is cloaking under Google’s and Bing’s rules, and the whole site gets penalised. For ChatGPT that would hit the main channel through the Bing index. Everyone sees the block, and so that it does not get in people’s way, on a phone 4 facts are visible with a “Show all facts” button, while the full text stays in the HTML.
Schema.org markup and llms.txt
Markup tells a robot directly: this is an organisation with this address and these profiles, this is a service at this price, these are questions and answers. We mark up the organisation (with directory profiles in the sameAs field), services with the lower price bound, FAQ, articles and cases.
The llms.txt file at the site root is a short map for AI: who we are, the main pages and facts about each. It does not replace the index, but it gives the assistant a ready table of contents. Ours is generated automatically on every site build, so new pages get into it at once.
Directories and reviews
AI assistants rely on independent sources, and this was our weak spot. In October 2026 we published profiles on Clutch, Sortlist and TechBehemoths; on Oct 19 DesignRush was added, and several more directories are in moderation. There are no reviews on these platforms yet; that is the next step. In Russian-language answers Alice quotes industry media (Sostav, Cossa, Adpass), so cases with figures are worth publishing in your market’s industry media too.
How to measure visibility in AI
A list of questions once a month
Write 20-30 questions the way a client would ask them, without your name: “recommend an agency for clinic advertising”, “how much does a property lead cost in Thailand”. Once a month ask them in ChatGPT with search on, Perplexity, Gemini and Copilot, and record:
- whether the company is mentioned;
- whether there is a link to the site;
- its position in the list;
- who else is named;
- which sources the assistant quotes.
Take the first measurement before the work starts, or there will be nothing to compare growth with. Our baseline on Oct 11, 2026 covered 32 questions. Alice named us in 4 of 31 recorded answers (13%) and linked to the site in 5. ChatGPT, Perplexity and Copilot could not be measured that day: without a login they do not answer, and under a personal account the owner’s memory and settings distort the answer. For an undistorted measurement you need a clean account or the API.
Leads from AI in your analytics
ChatGPT adds utm_source=chatgpt.com to its links. With it you can see visits and goals in GA4 or Yandex Metrica: on our site in August 2026, 17 tagged visits gave 5 contacts; in September, 23 and 5; in October (1-16), 24 and 10.
Some visits arrive without the tag, only with chatgpt.com as the referrer: 10, 9 and 1 visits in the same months for us. Add them to the segment, or the channel looks smaller than it is. Visits from mobile apps may arrive with no referrer at all and land in direct traffic, so check analytics against the CRM and ask clients how they found you.
The “AI” segment is built from the sources chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and other assistants. Goals: form submission, Telegram click, WhatsApp click. How we connect such goals to the CRM is on the analytics service page.
How long it takes
- Days to weeks: the index. Of 38 pages we published on Oct 11-12, 2026, Google had indexed 33 by the check on Oct 16 (Search Console, URL Inspection). The first impressions of new pages in Google usually come one to three weeks after indexing.
- Weeks: Bing. Bing is slower to take new pages without external links.
- Months: mentions in AI. The assistant needs the site in the index and other people writing about the company. Directories, reviews and media publications build up over months.
Nobody can guarantee a place in a ChatGPT answer: the model assembles the answer, and it changes from query to query. What can be guaranteed is the work and an undistorted measurement.
Where to start
- Connect Bing Webmaster Tools and Google Search Console and check how many pages are in the index.
- Set up an AI-source segment and contact goals in your analytics.
- Write 20-30 client questions and take a baseline measurement.
- Rewrite the main pages: a short answer and figures with a source, visible to everyone.
- Add organisation and service markup, an
llms.txtfile and profiles in three to five directories. - Repeat the measurement monthly and look at leads as well as mentions.
We do this work for our own website and now for clients: SEO and GEO from $750 a month, with a monthly AI visibility measurement and a report on organic leads. For a business website outside real estate, see SEO and GEO for business.
Sources
- Yandex Metrica, webnostra.pro, website segment: visits and contacts by source, August to Oct 16, 2026.
- Bing Webmaster Tools and Google Search Console (URL Inspection), webnostra.pro, checks of Oct 16, 2026.
- Our baseline AI visibility measurement, Oct 11, 2026, 32 questions.
- Directory profiles: Clutch, Sortlist, TechBehemoths, DesignRush.