How many Spaniards use ChatGPT to decide purchases: the data available in 2026
If you work at an agency, you've spent months citing (or hearing cited) figures on ChatGPT usage that nobody knows where they came from. "80% of young people already shop with AI", "half of all searches are on ChatGPT" — percentages that circulate on LinkedIn with no source, no date and no country. This article is the opposite: all the verifiable data that exists on Spain as of today, with its source beside it, and — just as important — an honest section on what no study has measured yet.
If you need ammunition for a proposal or a meeting, it's all here. If what you're citing isn't here, check the source before you use it.
How many people use AI in Spain: the two reference figures
Funcas, III Survey on AI (2026): frequent use of ChatGPT in Spain has gone from 4% to 28% in two years. It's the most cited figure and the most solid on ChatGPT specifically. Two readings worth keeping together: multiplying by seven in two years is an extraordinary adoption curve, and at the same time 72% of Spaniards still don't use ChatGPT frequently. Anyone who only counts one of the two halves is selling something.
INE: 37.9% of the Spanish population used generative AI in the last quarter of 2025. The most institutional figure on the list — the INE measuring generative AI in its official surveys is, in itself, a sign of the phenomenon's maturity. The figure is higher than the Funcas one because it measures something broader: any generative AI (not just ChatGPT) and any use during the quarter (not just frequent use).
In companies, adoption is ahead: 76% of Spanish SMEs use AI daily (II Hiscox Report, December 2025). Relevant to agencies on two counts: your clients already use AI to work, and whoever uses a tool daily soon starts using it to look for suppliers too.
How search is changing: the behaviour data
AI usage isn't simply adding to classic search without touching it; it's cannibalising it, and that's measured too:
- Zero-click searches have gone from 56% (May 2024) to 69% (May 2025) (data cited in stucom.com). Seven in ten searches end without a click to any website: the answer — increasingly often generated by AI — resolves the query on the results page itself.
- Where AI Overviews appear, the average organic CTR falls by 61% in Spain, and traffic from AI grows by 527% in a year (ismajimenez.com). Two faces of the same movement: the classic click shrinks while the new channel multiplies.
- Traffic arriving from AI answers converts at 14.2%, versus 2.8% for classic organic (roymo.es, citing sector data). Five times more. The visitor arriving after an AI answer comes filtered and with the decision half made — less volume, much more intent.
- Gartner forecasts 25% less traditional search volume before the end of 2026 (press release, February 2024). It's the only figure on this list that's a prediction and not a measurement; treat it as such. So far, the direction it pointed to is holding.
And when they ask, do companies appear?
Here comes the only large-scale study done on Spanish SMEs: at Surfeo we analysed 9,865 SMEs across 30 sectors in 10 cities, asking the 4 main AIs what a real customer would ask. The result: 91% appear in only 1 of the 4 AIs. With huge sector differences: among hotels, 1 in 4 is visible across several AIs; among garages, 1 in 50. The full data, by sector and city, is in the SME invisibility study.
The combined picture of all the data above: demand is growing (28% frequent use and rising), behaviour is changing (69% zero-click), the traffic AI generates converts five times more — and the vast majority of Spanish companies don't appear where that happens.
What is NOT known (and you should say out loud)
This section is missing from almost every article in the sector, and it's the one that'll give you credibility in front of a sceptical client:
There's no official figure for "purchases decided by ChatGPT" in Spain. Neither the INE, nor Funcas, nor any body yet measures how many purchase decisions are made inside a conversation with AI. What we have are indirect pieces — how many people use it, how the click is falling, how the traffic it generates converts — that point in the same direction, but the direct figure doesn't exist. Anyone giving you an exact percentage of "purchases decided by AI" is either inventing it or extrapolating a study from another country and another context.
Nor is it precisely known what share of ChatGPT usage is commercial (looking for suppliers, comparing products) versus work, study or leisure, nor how adoption breaks down by purchasing sector. And the behaviour figures move fast: any figure on this page has a date, and in twelve months several of them will be old.
Does that uncertainty mean it's too early to act? That's the underlying question, and it depends on what for: to promise guaranteed results, it'll always be too early; to measure and correct the sources, the adoption data already justifies it. The full argument, with the comparison every sceptic has in mind, is in is GEO another fad like NFTs?.
How to use this data in a meeting
Three rules that'll save you grief:
- Always cite the source and the year. "28% according to Funcas 2026" survives any follow-up question; "people already use ChatGPT for everything" dies at the first.
- Give both halves of the figure. The 28% frequent use and the 72% who don't. Paradoxically, acknowledging the limit of the data makes everything else you say more credible — and it protects you from the client who wants to cut channels that work, a conversation that has its own script in what to say when the client wants to cut Google Ads over AI.
- Close with the client's figure, not the country's. National statistics open the conversation; what closes it is seeing what the AIs say about their specific business. It's the difference between a headline and a diagnosis.
Frequently asked questions
Why does Funcas say 28% and the INE 37.9%? Which do I use?
They measure different things: Funcas measures frequent use of ChatGPT specifically; the INE measures any use of any generative AI during the quarter. Neither is wrong — use the Funcas one to talk about ChatGPT and habit, the INE one to talk about general AI adoption. Citing both, explaining the difference, is the option that reads best in a proposal.
Is there data by age or by sector?
The general surveys show that adoption is higher among young, urban audiences, but the fine cuts by purchasing sector don't yet exist with sufficient quality in Spain. For a specific client, the practical route is to flip the problem: instead of asking how many of their customers use AI, check what AI answers when asked about their category.
How often are these figures updated?
Funcas publishes its survey annually; the INE, in its ICT equipment and usage waves; the rest are one-off studies. Rule of thumb: don't use a 2025 figure in 2027 without checking whether there's a newer edition. In this market, a two-year-old figure is archaeology.
And how many of the clients in my portfolio are already affected?
That figure isn't in any survey: it's on their websites. Ask the AIs about each one — or run them through a free AI visibility test — and you'll have your own portfolio statistic in an afternoon. It usually resembles the study's 91% unsettlingly closely.
You've already got the country's data. The one that's missing is your client's: run the free AI visibility test and check which AIs they appear in — and which ones their competition answers in.
