The Purchase Decision Has Moved. Here Is Its New Address.

TL;DR

• ChatGPT has changed its nature in Belgium. Work-related messages fell from 47.1% to 26.8% of the total between July 2024 and June 2026 (OpenAI Signals v2.0, analyzed by Xavier Degraux, Aug. 2026). The rest is personal life.
• The #1 Belgian use is practical guidance, at 36.3%: how to do something, what to choose, how to decide. Higher than the global average of 32.1%.
• An estimated 3.5 million Belgians use ChatGPT (GWI via Trends Tendances, 2026). Their first reflex is no longer to search. It's to ask for advice: in the evening, at home, in a conversation.
• The new address of the purchase decision is a generated answer, and the handful of sources it cites. That position is built on your site and, above all, beyond it. And in Belgium, it is still largely up for grabs.

What do the Belgian OpenAI Signals numbers actually show?

In two years, ChatGPT in Belgium went from a work tool to a life tool. On August 6, 2026, OpenAI published version 2.0 of Signals, its public dataset on individual ChatGPT usage: 300,000 messages sampled per month, Free, Go, Plus and Pro accounts, covering July 2024 to June 2026, filterable country by country. Belgian consultant Xavier Degraux ran the Belgian series through the wringer, and his analysis (August 2026) deserves the stop.

The headline movements:

Source: OpenAI Signals v2.0, Belgian series analyzed by Xavier Degraux, August 2026.

The tipping wasn't gradual. The work share eroded slowly for twelve months, then dropped in two: from roughly 37% in May 2025 to roughly 25% in July 2025, stabilizing since. And Belgium isn't an outlier: it tracks the global curve (51.3% → 30.4%), just 3.6 points below it, with a distinctly more utilitarian profile: more practical guidance, more information seeking, less personal expression and multimedia than the world average.

Why should brands care about a category called "practical guidance"?

Because "practical guidance" is where decisions form. Read what the category actually covers in OpenAI's data dictionary: how-to requests, tutoring and teaching, creative ideation, and health, fitness and self-care. In plain language: how do I do this, what should I choose, how do I decide.

Now add the intent data. In Belgian non-work messages, "asking" (requesting information or advice) dominates at 48.7%, above the global 45.1% (OpenAI Signals, 2026). Practical guidance alone represents 40.4% of Belgian personal conversations, versus 35% worldwide. Two categories (practical guidance and information seeking) account for nearly two thirds of everything Belgians say to ChatGPT outside work.

Reread that slowly.

The first reflex of an estimated 3.5 million Belgians (37% of the 16+ population, per Global Web Index figures for H1 2026 cited by Bernard Cools of Space in Trends Tendances) is no longer to search. It's to ask for advice. In the evening. At home. In a conversation. Not a list of ten blue links to sort through: a synthesized recommendation, plus the handful of sources the engine chose to cite.

That is the new address of the purchase decision: a generated answer, and the few sources behind it. If your brand isn't in the answer or in those sources, you are not at the address where the decision is being made. You're at the old one.

What does this data not say? (The part most commentary skips)

Honest reading is part of authority, so here are the limits, clearly stated. Degraux's analysis is exemplary on this point, and it deserves to be carried over rather than sanded off:

• Signals publishes shares, not volumes. A declining share of work messages inside a growing total is fully compatible with a rising absolute number of work messages. The safest reading: personal use grew much faster.
• The 26.8% work figure is a floor, not a measure. Enterprise and Codex usage is excluded from the sample; OpenAI says so explicitly. Professional AI usage in Belgium is higher than this dataset can show.
• It's ChatGPT only. Nothing here covers Gemini, Copilot, Claude, Perplexity or Le Chat, engines capturing a growing share of Belgian usage, each with its own citation behavior. A single-engine read is never the full picture.
• Classification is automated and deliberately noised for differential privacy. These are solid orders of magnitude, not audience measurements.

None of these caveats weakens the strategic signal. They sharpen it: what the data reliably shows is a structural migration of everyday advice-seeking (the raw material of consumer decisions) into AI conversations. The direction is unambiguous even where the decimals are soft.

Does the rest of the research point the same way?

Yes. From three independent angles.

The buyer side: AI narrows, search validates

Semrush's survey of 519 US B2B decision-makers (March–April 2026) found that 41% now start their vendor research with AI and validate via search, and that after an AI recommendation, 71% visit the vendor's website and 63% re-Google the name. On the consumer side, Semrush's study of 1,030 US shoppers (fielded Dec. 2025) found 55% already use AI tools weekly for product research. The pattern is the same in both worlds: AI opens the decision; verification follows. We unpacked that sequence (and why authority now beats awareness inside AI answers) in our analysis of the Semrush B2B data.

The engine side: commercial questions pull in third-party sources

Kevin Indig's research (Growth Memo) adds the mechanism that matters most for brands: the higher the purchase intent behind a prompt, the more the engine's answer leans on third-party sources (review platforms, communities, publishers) rather than brand domains. Advice-type questions are exactly where engines triangulate across independent sources. Which is why 85% of brand mentions in AI answers come from third-party pages, not the brand's own domain (AirOps, 2025).

The behavior side: answers increasingly end the journey

Bain & Company found that 80% of consumers rely on zero-click results in at least 40% of their AI searches. When the answer satisfies, the conversation is the destination. The brands present inside that answer capture the decision; the brands waiting one click further away often never see the visitor.

How does a brand move to the new address?

You don't buy your way into a generated answer. You build your way in, on both sides of the signal.

On-site: answer the questions people actually ask at 9 PM

• Publish for "how to choose", not just "why us". The Belgian data says the demand is how-to, what-to-choose, how-to-decide. Generative engines cite the content that serves that demand: guides, comparisons, structured FAQs, decision frameworks, the kind of content that is useful without mentioning your brand. Commercial pages are near-invisible in these answers.
• Make it retrieval-ready. Self-contained sections, front-loaded claims, one question per chunk. 44.2% of LLM citations come from the first 30% of a piece of content (Growth Memo, Feb. 2026).
• Mind the Belgian specificity: three languages, one entity. 61% of Belgian ChatGPT users are Dutch-speaking (GWI via Trends, 2026). Your entity description must stay verbatim-consistent across FR, NL and EN surfaces; variation across languages fragments how engines recognize you.

Off-site: be corroborated where the engines triangulate

• Independent validation outranks self-declaration. Brands are 6.5x more likely to be cited via external sources than via their own content (Onely, 2025). Reviews, sector press, expert mentions, comparison articles: the sources an engine cites for advice questions are overwhelmingly third-party.
• Prioritize the levers with proven impact. The three AEO levers with the highest business impact are all off-site: Digital PR (10/10), Guest Posting (10/10), Social Media References (9/10) (NP Digital, survey of 100 marketers, Jan. 2026).
• Quality of corroboration over quantity. Three strong, thematically relevant sources that tell the same story about you beat fifty generic directory listings.

A pattern from the GEO monitoring we run daily at PingPrime across ChatGPT, Google AI Mode and AI Overviews: for advice-type and "how do I choose" questions, the sources engines pull are educational guides, comparison pages, official documentation and video explainers, third-party far more often than brand-owned, and homepages almost never. Our monitoring keeps confirming what the Belgian usage data implies: the engines answer advice questions with educators, not advertisers.

The window is Belgian, and it is still open

Here is the asymmetry worth acting on. The demand side has already moved: 3.5 million Belgians, practical guidance first, decisions forming in conversations. The supply side hasn't: most Belgian brands still optimize for the old address of rankings, impressions, and the click that comes after the decision instead of the answer where it forms.

That gap is the opportunity. Narrative authority at the new address isn't rented monthly like media space. It's infrastructure (built signal by signal, source by source, over months) and it compounds. The brands that build it now, while the Belgian answer space is still thinly contested, will be the references the engines cite, trust and repeat when everyone else shows up.

The decision has moved. The question is no longer "do they find us when they search?"

It's "are we in the answer when they ask?"

FAQ

What is the number one use of ChatGPT in Belgium?

Practical guidance (how-to requests, tutoring, creative ideation, health and self-care) at 36.3% of Belgian messages in June 2026, above the global 32.1% (OpenAI Signals v2.0, analyzed by Xavier Degraux, 2026). Writing (21.9%) and information seeking (20.4%) follow.

How many Belgians use ChatGPT?

Around 3.5 million, or 37% of Belgians aged 16+, per Global Web Index H1 2026 figures cited in Trends Tendances, of whom 61% are Dutch-speaking. OpenAI Signals itself publishes message shares, not user counts.

Has professional ChatGPT use declined in Belgium?

Its share of messages fell from 47.1% to 26.8% between July 2024 and June 2026, but this is a share, not a volume, and a floor: Enterprise and Codex usage is excluded from the sample. The reliable reading is that personal use grew much faster.

How does a brand get into AI-generated advice?

By building both sides of the signal: on-site, educational retrieval-ready content answering how-to and what-to-choose questions in every relevant language; off-site, independent corroboration from reviews, press and comparisons, since 85% of brand mentions in AI answers come from third-party sources (AirOps, 2025).

------

HOW THIS WAS WRITTEN: Topic, angle, convictions: mine. AI assists the writing. I make the call.

Sabrina Bulteau is the co-founder of PingPrime.ai, GEO Expert and specialist in Narrative Authority in AI Search. She helps brands, institutions and media become the reference AI engines trust, cite, and repeat (not just one option among many) by working both sides of the signal: on-site (structure, narratives, architecture) and off-site (earned media, platforms, co-citations, sector press). Co-founder of Be Connect (acquired by iO Group) and Sench, active within CEC Belgium, she brings 25 years of experience in digital growth and strategic positioning.

‍