The Hidden Rules ChatGPT Uses to Pick Brands (2026)

By Vynus Designs · Published 21 May 2026 · Updated 1 July 2026 · 7 min read

ChatGPT picks which brands to recommend using seven signals: source authority, topical match, content structure, freshness, schema markup, brand entity strength and off-site mentions. Referring domains carry around 30% of the weight and brand search volume another 25%. Hit all seven and ChatGPT starts citing you inside 8 to 12 weeks.

ChatGPT reached an estimated 1 billion weekly active users by May 2026 and now handles roughly 2.5 billion prompts a day. A growing share of those prompts are buyer questions: “best web designer in London,” “Greek restaurant near me,” “agency that does branding and SEO.” ChatGPT answers with a shortlist of brands, every time. And while ChatGPT still leads, it no longer owns the market: the AI chatbot market share 2026 numbers show Gemini and Claude taking a fast-growing slice of those buyer questions.

Most business owners assume that shortlist is random. It is not. There is a clear set of signals ChatGPT uses to decide which brands it names and they are measurable. The brands cited consistently across UK, Greek and international queries are the ones quietly hitting all seven.

This is what we look for first when a Vynus client asks how ChatGPT picks which brands to recommend in their category. Seven signals, in priority order, with the lever for each.

How ChatGPT actually picks which brands to recommend

ChatGPT does not just guess. It combines what it learned during training with a live retrieval step (the part that reads the web in real time) and then runs the candidate sources through a set of trust and relevance filters before naming a brand in an answer.

A peer-reviewed analysis of 11,000 SearchGPT queries published last year identified the same handful of factors deciding who gets cited. Independent studies have since confirmed the weighting. The big four are source authority (referring domains), brand entity strength (how often the brand is searched and discussed), topical depth on the page itself and content freshness inside the last 30 days. Three more (schema markup, off-site community mentions and crawl access) act as the on or off switches that decide whether the page even enters the candidate pool.

The good news for UK and Greek small businesses is that none of these signals require a big budget. They require the right sequence.

The 7 hidden signals that decide who gets cited

Ranked by how strongly they correlate with citation in our audits, with the supporting research alongside:

1. Referring domainsThe clearest single signal. Roughly 30% of citation weight comes from how many trusted sites link to you. One quote in an industry publication beats ten posts on your own blog.
2. Brand entity strengthAround 25% of the weight. ChatGPT measures how often your brand name is searched, mentioned and described consistently across the open web. Consistent positioning matters more than volume here.
3. Community signalsRoughly 20% of the weight. Reddit threads, Quora answers and YouTube mentions punch above their weight. YouTube specifically correlates with citation at 0.737, the single strongest signal measured in the study.
4. Topical match and depthPages with question-shaped headings, direct one-sentence answers and FAQ blocks get extracted into AI answers far more often than narrative marketing copy. AI engines pull the cleanest passage, not the prettiest.
5. FreshnessContent updated inside the last 30 days picks up roughly 3.2 times more citations than older material. Visible Published and Updated dates matter here as well.
6. Schema markupPages with structured data are cited at roughly three times the rate of pages without it. Organization, LocalBusiness and FAQPage schema do most of the work.
7. Crawl accessThe on or off switch. OpenAI runs three separate crawlers (GPTBot for training, OAI-SearchBot for ChatGPT search and ChatGPT-User for live user fetches). Each one is controlled independently in robots.txt. If you copied a “block all AI bots” template from 2023, you may have switched yourself off without realising.

How to send those 7 signals: the Vynus playbook

We run this list in order. Items higher up unlock the ones below, so do not skip ahead.

1. Open robots.txt and explicitly allow OAI-SearchBot, GPTBot, ChatGPT-User, Google-Extended, PerplexityBot and ClaudeBot. Block only the bots you genuinely do not want crawling your site.
2. Add an llms.txt file to your domain root. Start with an H1 site name, one-line summary, then a short list of your most important pages with one-line descriptions each. Keep it under 50 lines.
3. Add Organization, LocalBusiness (for UK and Greek SMEs with a physical presence) and FAQPage schema across the site. Use a JSON-LD @graph block per page and verify with Google’s Rich Results Test.
4. Restructure your top five revenue pages. Make every H2 a real question. Put a direct answer in the first 40 to 60 words of each section. Add a five-question FAQ to each. This is the single change that moves the needle hardest.
5. Date-stamp the content. Visible Published and Updated dates on every page. Refresh your top five pages at least every 30 days.
6. Build off-site signals deliberately. One quote in an industry publication, one substantive Reddit thread and one short YouTube video per month. Pace beats intensity.
7. Get mobile Core Web Vitals into the green. If GPTBot times out on your pages, none of the above matters. PageSpeed Insights gives you the priority list.

If you would rather not run all seven yourself, this is the exact work covered in our SEO and AI visibility services.

What this looks like for UK and Greek small businesses

The UK is unusually behind on this. The Department for Science, Innovation and Technology (DSIT) reported in January 2026 that just 16% of UK businesses use AI in any structured way, against 88% globally. Greece sits in a similar position. For Vynus clients that gap is the opportunity. Most local competitors have not done any of the seven steps above.

When we audited a Greek hospitality client in north London last quarter, an old security plugin was silently blocking OAI-SearchBot in robots.txt. Twenty minutes of corrections plus a fresh llms.txt later, ChatGPT was citing their tasting menu page within four weeks. That is the standard timeline we see when a business has strong product but weak technical foundations.

For multilingual sites (English plus Greek or English plus another EU language), each language version needs its own schema, dates and contact details. Do not rely on auto-translated content for AI Overviews. Translate it properly or skip the second language until you can.

How long until ChatGPT starts recommending you?

Measurable citation changes typically begin 8 to 12 weeks after the playbook is implemented. Sustained AI visibility takes three to six months of consistent work. Two things accelerate the curve.

First, frequency. Sites publishing one substantive article a week see citation growth roughly twice as fast as monthly publishers, because LLM retrieval indexes weight freshness heavily. Second, distribution. An article that picks up one earned link in its first week tends to start collecting citations two to four weeks earlier than the same article published without any outreach.

If you are starting from zero, expect the first ChatGPT citation around weeks six to eight. Anything earlier is a bonus. Do not change the strategy before that point.

Frequently asked questions

Sources

1. Overview of OpenAI Crawlers · OpenAI Developer Platform, 2026
2. The /llms.txt file specification · llmstxt.org
3. Sam Altman says ChatGPT has hit 800M weekly active users · TechCrunch, October 2025
4. UK AI adoption research · Department for Science, Innovation and Technology, January 2026
5. Google Rich Results Test · Google Search Central
6. How to rank in ChatGPT (citation factor analysis) · Wellows, 2026