Siavak Analytics

Blog

Google searches are getting longer. That's where the shortlist splits.

People have started typing into Google the way they talk to an AI assistant. Not “best running shoes”, but “running shoes for flat feet, under 120, for a half marathon in March”. The buyer is putting their situation into the question.

That matters more than the traffic numbers that usually lead these reports. A short question gets the category’s shortlist. A long question gets the buyer’s own shortlist, and that’s where brands win or lose without seeing it.

A short question gets the category's shortlist. A long one gets the buyer's.

Longer questions reached Google

The new State of Search report from Semrush and Datos tracks what people do on desktop in the US and Europe. Its most interesting chart this quarter is also its smallest number: how many Google searches run past 15 words.

Searches of 15+ words hit a record on Google in September.

Share of Google desktop searches with more than 15 words.

USA

May 2026, the earlier peak

1.82%

August 2026

1.67%

September 2026

2.58%

Europe, including the UK

September 2025

0.33%

July 2026

0.71%

September 2026

1.32%
A small share, moving fast. In Europe it is four times what it was a year earlier. Searches of 9 to 14 words reached new highs too, and 5 word searches fell to a low.

Semrush and Datos, "State of Search Q3 2026", desktop clickstream panel, US and Europe including the UK.

Rand Fishkin, who comments on the report, called it the first real break in a number he watches closely. His reading:

"People are treating Google search more like an AI assistant."

Rand Fishkin, State of Search Q3 2026

They have found it can give them similar answers, so they ask it the same way.

The share is still small, and desktop only. But the direction is the point. The way people ask an assistant is now how a growing slice of people ask Google, whether the answer comes back as links, an AI Overview or AI Mode.

AI tools grew, and search held

None of this means people left Google. AI tools grew by about four fifths in a year and still take a fraction of the desktop time that search does.

The share of desktop web activity that went to AI tools, September 2025 and September 2026.

September 2025September 2026

USA

0.72%
1.29%

Europe

0.86%
1.42%

Semrush and Datos, Q3 2026, appendix. Traditional search was 9.47% in the US and 9.95% in Europe in September 2026.

Google’s AI Mode grew too, to 0.31% of US desktop activity in September and 0.43% in Europe. Rand’s warning on it is worth keeping in mind:

"AI Mode is, like other AI assistants, far less friendly about sending traffic out."

Rand Fishkin, State of Search Q3 2026

So the change is less about where people ask and more about how they ask. Rand’s own summary of the quarter is that it looks far more stable than the “AI will take over everything” headlines suggest. The question got longer in both places.

Why a long question changes the answer

Here is what we see in our own research. When the buyer says who it’s for, what they can spend or what matters to them, the AI assistants change the list. In almost every brand we have studied, one buyer situation flipped the result.

One watch brand from our research: how often the AI assistants put it first, by the buyer in the question.

Shopping for a gift

52%

Wants it made in America

18%

Siavak Analytics Research, October 2026. Brand anonymised.

The same watch was the assistants’ first pick for gift buyers about half the time. When the buyer wanted something made in America, it fell to fewer than one answer in five. Another brand, a donation pickup service, was named in every answer about a move with only bags left to give, and in none once the buyer mentioned furniture.

The brand that comes first on the short question can drop out once the buyer says who it's for.

A brand that checks itself on the short question sees the category answer. Its buyers, more and more, get a different one.

Then they check

The report has one more finding that fits. After using an AI tool, the site people visit next most often is Google, in the US and in Europe. Wynter reports the same from the other end: of 200 B2B buyers who researched a vendor with AI, 72% went to vendor websites right after, and the less sure the AI left them, the more they checked.

The website confirms the shortlist. The answer makes it.

So the buyer arrives with a shortlist and an opinion. The website and the search result can confirm what the assistant said or contradict it, but the shortlist was made before the visit.

What it means for a brand

Ranking for the keyword tells a brand less every quarter about the answer its buyers actually get. The useful questions are different:

  • Which buyer situations does the brand win, and which does it lose?
  • What does the assistant say about it when the buyer adds a budget, a need or a doubt?
  • Which competitor takes its place when it drops out, and why?

None of that shows in traffic or rankings. It shows in the answers, buyer by buyer.

We research what the AI assistants tell buyers about a brand and its competitors, from the short question to the specific one. If you want it for yours, you can request a free assessment.

Sources: Semrush and Datos, "State of Search Q3 2026: Behaviors, Trends, and Clicks Across the USA & Europe", with commentary by Rand Fishkin, November 2026 (desktop clickstream, July 2025 to September 2026), shared by Rand Fishkin on LinkedIn, 5 November 2026. Wynter, survey of 200 B2B buyers, shared by Peep Laja, 5 November 2026. Our own figures come from Siavak Analytics Research, September and October 2026, brands anonymised.