Playbook
ChatGPT and Google AI Overviews: two engines, two jobs
A buyer abroad can ask the same question in ChatGPT or in Google and get two different answers, built from two different sets of sources. Kaan Gülten's talks to Turkish exporters make one point repeatedly: there is no single optimisation that works for every engine. This page sets out what each engine reads, who you are likely to meet on each, what both need from you, and why they have to be measured separately.
By Kaan Gülten · Based on talks given in Turkish · Published · 5 min read
There is no single optimisation for every engine
In his IGEXX 2026 keynote, Kaan Gülten compared three engines side by side: Claude, ChatGPT and Google's SI answers. Each relies on a different set of sources. As presented on the slide, ChatGPT shares only 8% of its sources with Claude. A company that is visible in one engine can be absent from another.
Brantial's own research points the same way. In its women's fashion report (August 2026; 1,000 unbranded queries across three models; 1,695 brands), 74.7% of the brands that appeared did so in only one of the three models, and 11.5% appeared in all three. This is a commercial company's research, read together with its question set, model choice and period; it is not a probability of future visibility.
Side by side
| ChatGPT | Google AI Mode and AI Overviews | |
|---|---|---|
| How the talk describes it | The engine everyone uses | Where the volume is |
| When it looks at the web | On almost every question | Not described on the slide |
| What it values | Reddit, Wikipedia and user reviews | Pages it can split into parts; sources that say the same thing, grouped into the answer |
| What it needs from you | To be talked about on the platforms it reads | No ambiguity: the company name and the claim in the same sentence |
| Crawler to allow | OAI-SearchBot (search index) and ChatGPT-User (reading on a user's request) | Googlebot (serves Search and AI Overviews) |
| First-party report | Not mentioned in the talks; measured with controlled, repeated questions | Google's Generative AI performance report in Search Console (AI Overviews and AI Mode impressions) |
Claude, the third engine on the slide, is described as the corporate buyer's engine: as presented on the slide, it takes 79% of its sources from the top ten results of the Brave search engine, looks at the web for questions about the current year, rankings and comparisons, rarely looks at forums, and prefers guides and list articles.
What both engines need
The common ground is the same in every engine on the slide: clear, numerical, sourced and current information.
“So-and-so Textiles is an İzmir-based manufacturer that exports 5 million metres of denim a year to Europe.”
The New Digital Customer, slide 9, IGEXX 2026 · translated from Turkish (placeholder company name)
Before either engine can use it, the information an agent needs has to be readable by a machine:
- Company: year founded, capacity, certificates, export countries, reference sectors.
- Product: technical specification, minimum order, lead time, payment terms, price range and current stock.
- Proof: reviews in English and German, audit report, factory tour, executives' own writing.
Language matters too. In Kaan Gülten's view, SI systems know less in languages other than English, so a Turkish company that publishes German, Italian and Arabic content faces little competition in those languages.
Who you meet on each
Kaan Gülten's reading is that the two channels do not reach the same people, and that the audience moving to conversational assistants is not a random half of the market.
“If I'm targeting my mother and father, I need to use Google. But if it's an audience that adapts quickly to new technology, with a somewhat higher income and fast decisions, chat platforms will directly affect your sales.”
The Change in Search, slide 4, Ankara E-Export Summit 2025, from the recorded talk · translated from Turkish
“If half the traffic goes, what happens? The two halves of your audience won't move there evenly. People with certain behaviours, a certain level of education and a certain income will go.”
The Change in Search, slide 5, Ankara E-Export Summit 2025, from the recorded talk · translated from Turkish (shortened; the full sentence is on the slide page)
The size of a channel is a separate question from whether your buyer is in it.
“‘There is a large user base’ and ‘my buyer buys through this channel’ are different sentences.”
Branding and Marketing in the Age of SI, slide 112, UİB Academy 2026 · translated from Turkish
You learn the second from customer conversations and sales records, not from platform user counts.
Measure them separately
Kaan Gülten separates measurement into four layers, each answering a different question:
- Platform visibility. Google's Generative AI performance report in Search Console shows AI Overviews and AI Mode impressions; Bing has its own AI Performance report. Numbers from the two platforms are never added together.
- Controlled answer samples. Fixed questions, asked repeatedly. This is how ChatGPT, Gemini, Perplexity and Claude are measured: is the brand mentioned, and is the information correct?
- Site behaviour. Analytics and quote form events: visits and actions that can be identified.
- Commercial result. CRM and sales records: qualified requests, quotes, won business.
Do not squeeze them into a single score. Part of the effect will not show up anywhere: a buyer who saw your name in ChatGPT and visited two days later appears in analytics as “direct traffic”. Asking “Where did you first hear about us?” on the quote form, with “SI assistant” as an option, recovers some of it.
Measure each market in its own language and from that market. In a fictional case from the UİB training, the same company appeared in more than half of the answers when asked in English from Turkey, and almost never when asked in German from Germany, because it had no German technical page and German answers cited German sector portals. Telling a model “answer for Germany” is not the same as a test by a real user in Germany. For every test, record the question, target country, query language, actual test location, platform, whether web search was on, the date, whether it was a new chat, and the sources. Do not treat API results and consumer interface results as one series.
What may change next
Two of Kaan Gülten's published forecasts concern exactly this comparison. Neither gives a date, and neither has been assessed yet:
- Google's classic search will give way to Gemini (said in 2025): competition will push Google to make Gemini the main experience on google.com itself, beyond adding a few SI features to the results page.
- Ads in SI answers could change the balance of competition (said in 2025): if SI platforms start offering paid placements, budgets gain a new lever next to being recommended organically.
The UİB training already advises measuring organic and paid results separately: OpenAI's Sponsored Agents test (16 September 2026, United States) is presented apart from the independent answer.