The click isn't dead: it's out for a drink

Two stories circulate about web traffic. One holds that the audience is moving to artificial intelligence and that nothing is wrong: just buy some new tactics (GEO, SEO+ and whatever gets invented next month) and keep getting by. The other version tells us the web is over and there is nothing left to scrape from it. This year’s data contradicts both positions. And what’s left in between is considerably more demanding. The click hasn’t disappeared: it’s become competitive.

What’s been lost, like tears in the rain of pixels

It’s worth starting with the number that proves the pessimists right, because it’s real, and in Spain it’s especially harsh.

SE Ranking has tracked 101,574 websites with valid analytics between January 2025 and April 2026. According to that analysis, organic search went from contributing 52.02% of traffic on Spanish websites to contributing 40.70%, a drop of 11.32 percentage points in just over a year. The global average fell too, but by much less: 2.59 points, from 45.34% to 42.75%.

That we’re moving four times faster than the rest of the world is less mysterious than it looks. In 2025 we depended on organic search almost 7 points more than the global average, and in 2026 we’re already 2 points below it. Nothing different has happened to us than to everyone else: we simply had more to lose, and we lost it sooner.

The interesting question is where that traffic went, because if the migration story were true, eleven points would have to show up somewhere. The same source measures traffic coming from AI assistants, and in Spain it went from 0.23% to 0.30% of the total. Seven hundredths of a point. In the global average, from 0.24% to 0.32%: eight hundredths, practically the same. Whatever is happening, it isn’t a transfer, and it isn’t a Spanish anomaly.

Here it’s worth switching off an intuition we carry almost without noticing. Energy follows the rule that it’s neither created nor destroyed, only transformed, and we tend to think of traffic the same way: if I lose it, someone else gains it, and all I need to do is find out who, to go looking for it there. Web traffic doesn’t work like that. And it’s worth being precise about what the figure says: SE Ranking measures the share that search represents within each website’s traffic mix, not the number of visits. Organic search losing eleven points of share doesn’t mean eleven points’ worth of visits have moved elsewhere; it means search carries less weight in the mix. And they haven’t turned up at the place they supposedly moved to.

−11.32 pp

organic search's share of traffic to Spanish websites has fallen: from 52.02% to 40.70% in just over a year

SE Ranking · 101,574 sites, 2025-2026

−2.59 pp

the global average of that same share has fallen: from 45.34% to 42.75%

SE Ranking · averages across 250 countries

+0.07 pp

traffic from AI assistants in Spain has risen: from 0.23% to 0.30% of the total

SE Ranking · 2025-2026

+0.08 pp

the global average of AI-driven traffic has risen: from 0.24% to 0.32%

SE Ranking · 2025-2026

Whatever is happening, it isn't a transfer: they haven't turned up at the place they supposedly moved to.

What hasn’t been lost: it was out on the town

If we stopped here, the conclusion would be the blackout story. But there’s a second measurement, just as solid, that looks at the same phenomenon from the other end and tells a different story.

The State of Search Q1 2026 report from Datos is built on a desktop clickstream panel across the United States, the EU and the UK. Instead of measuring where a website’s traffic comes from, it measures what people do inside the search engine. And in Europe and the UK, searches that end without any click fell from 22.5% in December to 19.6% in March, while clicks to organic results rose from 45.6% to 46.0%.

The same thing happens in the United States: no-click searches fall from 24.5% to 22.4% and organic clicks rise from 42.0% to 44.9%. And the rest of the picture points in the same direction on both sides of the Atlantic. AI tools account for 1.08% of visits in Europe and 0.93% in the United States, always below 2% despite all the noise, while Google holds 95.5% and 94.3% of desktop search. The search engine isn’t sending fewer clicks: it’s sending a few more.

19.6%.6%

of searches in the EU and UK end without any click; it was 22.5% in December. In the US: 22.4%, down from 24.5%

Datos · State of Search Q1 2026

46.0%.0%

of European searches end in a click to an organic result; up from 45.6%. In the US: 44.9%, up from 42.0%

Datos · State of Search Q1 2026

1.08%.08%

of visits in Europe go to AI tools (0.93% in the US): below 2% despite all the noise

Datos · State of Search Q1 2026

95.5%.5%

of desktop search in Europe is still Google's; in the US, 94.3%, recovering from 92.1% in December

Datos · State of Search Q1 2026

The search engine isn't sending fewer clicks: it's sending a few more.

Everything everywhere all at once

Placed side by side, the two studies look incompatible, and they aren’t: they measure different things. SE Ranking measures a website’s traffic mix, that is, what portion of your visits arrives through search. Datos measures behaviour inside the search engine, that is, how many searches end in a click. Nothing stops the search engine from handing out more clicks while a smaller portion of the total reaches your website.

Before drawing conclusions, two limits are worth stating. Each study defines a “no-click” search its own way (some count a click that stays within Google’s own properties, others don’t), so percentages from different sources aren’t comparable with each other. And the Datos panel is desktop only, which leaves out roughly two-thirds of searches, the ones that happen on mobile; the report itself flags this.

With those caveats in place, the combined reading is the whole point: if the same clicks or more are being handed out and a smaller portion reaches you, those clicks haven’t evaporated, they’ve changed hands. The click hasn’t disappeared; it’s become competitive.

If the same clicks or more are being handed out and a smaller portion reaches you, those clicks haven't evaporated, they've changed hands.

From there follows the only question that matters: what makes a website earn that click today.

Three “readers” for the same website

The answer has changed, and not because new channels have appeared, but because new readers have. Your website now has three at once, and all three have become more demanding.

The first is still a person, only they arrive less often and with clearer ideas. Datos observes that 6-to-8-word queries keep growing steadily without short ones dropping (people aren’t searching differently, they’re searching more precisely) and that informational intent rises to 60.7% on Google Europe. Whoever arrives already knows enough and has come to resolve something specific, so they decide in seconds whether this is the page they were looking for.

The second is the search engine, which still sends the bulk of the traffic and still rewards roughly the same things it always has. What’s changed isn’t its criteria, but that the bar is higher because there are more candidates for every click.

The third is new: a language model that comes in, reads, extracts what it needs and leaves without registering a visit, but carries your name into a conversation you’re not part of. When Cyrus Shepard aggregated 54 studies, patents and experiments to understand what determines whether a model cites a source, the highest-rated factor wasn’t the content but something far more mundane: whether the URL is accessible. Second was classic organic ranking, ahead of domain authority, structured markup and any tactic sold as the latest thing.

And it’s also the clumsiest of the three. An analysis of ChatGPT’s retrieval stack published in Search Engine Land last August, based on 1,200 of the model’s answers, spells it out without mercy: it doesn’t execute JavaScript, so whatever is rendered client-side simply doesn’t exist; it keeps the text and discards the rest; it anchors on the headline and whatever it finds around it; and if a page is over four megabytes it doesn’t trim it: it discards the whole thing and reads nothing. You don’t need to be technical to recognise the list: it’s the usual one from any accessibility audit, under a different name.

They all want the same thing

The convenient part of having three readers is that, in practice, they ask for almost the same thing, so serving all three doesn’t mean tripling the effort; it means doing well a job we already knew. There’s no need to build three websites.

A clear content structure helps a person find their way, helps the search engine understand the hierarchy, and helps the model extract the right fragment. A page that loads fast and doesn’t depend on JavaScript to show what matters reads just as well on a phone with poor coverage, to a crawler, and to an assistant. A coherent information architecture, with no sections contradicting each other, keeps the user from getting lost and the model from reasoning badly about you. And semantic markup and accessibility stop being the legal requirement ticked off at the end and become the layer that makes your content interpretable by anything that reads it, with eyes or without them.

What doesn’t work is the opposite shortcut: writing for machines. Studies that have tested rewriting content “optimised for AI” under real conditions (C-SEO Bench, presented at NeurIPS 2025, and SAGEO Arena) find that retrieval gets worse, because the text ends up so far from the natural language of real queries that the system stops finding it. And of the llms.txt files published last year, Ahrefs found across 137,210 domains that 97% have never been read by a single bot. Optimising the last step of a funnel you no longer reach is expensive for changing nothing.

What this means if your website doesn’t live off volume

At Digital Bakers we work with organisations (hospitals, universities, ports and industry…) that have almost never lived off traffic volume and that need something else: that when someone asks about a treatment, a training programme or a procedure, the answer gets built from their information, not from a forum’s.

For them the change is more favourable than it looks, because they have exactly what’s scarce: first-hand knowledge, accredited authority and information no one else can give the same way. The problem is rarely the content; it’s usually that the website holding it isn’t built for that knowledge to be found, read and cited.

Our position, with our hands in the dough

Nobody is going to bring back the traffic that went out for a drink just by optimising your website for AI, and we don’t say that on intuition: the data we’ve reviewed says it.

What we maintain is more boring and considerably more useful: if every click costs more to earn, the visit that arrives is worth more than ever.

But most websites are still built for the abundant visitor of years gone by. Our job is to make sure that whoever comes in (a person, a crawler or a model) finds an impeccable, efficient experience: clear structure, fast loading, real accessibility and a narrative that answers the question that brought someone there.

In short: the click is still there, but it no longer comes on its own. It used to be enough to show up; now you have to earn it. And that success depends, no more and no less, on your house being in order and on what’s inside being worth the trip. There’s also a part of all this that reconciles me quite a bit with the moment: people have been saying for a while that artificial intelligence will make us lazier, and in our trade, for now, it’s doing the opposite. By raising the bar for what anyone can find on any page, it forces yours to be better thought out, better written and easier to read. Not a bad deal. It’s the same as ever, only done better.

JM Sanchón, Content Director at Digital Bakers

Sources

• SE Ranking, AI Traffic in Spain. 101,574 websites with valid analytics tracking, January 2025 to April 2026; global averages calculated across 250 countries.
• Datos, State of Search Q1 2026: Behaviors, Trends and Clicks Across the US & Europe. Desktop clickstream panel across the United States, the EU and the UK.
• Zyppy (Cyrus Shepard), aggregation of 54 studies, patents and experiments on citation factors in AI engines, May 2026.
• Search Engine Land, “Inside ChatGPT’s retrieval stack: the index, cache, and pages it actually reads”, Olivier de Segonzac, 17 August 2026, on RESONEO’s analysis of 1,200 ChatGPT answers.
• C-SEO Bench (NeurIPS 2025) and SAGEO Arena (2026), on the real effect of optimisation tactics for generative engines.
• Ahrefs, study on llms.txt. 137,210 domains, server logs from May 2026.