News: On September 11, 2026, Jason Tabeling, Head of Solutions at Further, published an analysis of Google Ads query data on Search Engine Land covering January 2025 to August 2026: the impression share of 1-to-2-word queries fell from 42% to 24%, while 3-to-4-word queries climbed from 33% to 48% (Search Engine Land, September 11, 2026).
No algorithm moved, and that is exactly what makes this data interesting. Nobody changed the ranking: people changed the way they type. In nineteen months, the query segment most keyword plans are built on lost nearly half its weight. The numbers come from Google Ads, but the behaviour they measure happens in the same search box that feeds your organic traffic.
What the numbers actually say
Impression share is the percentage of total ad displays captured by a given query segment; conversion share applies the same calculation to the actions completed after the click. The analysis compares query distribution by length between January 2025 and August 2026. Two movements mirror each other. On visibility, short queries fade: 18 percentage points of impression share lost in absolute terms. On value, the 3-to-4-word block does not merely gain volume, it takes the conversions with it: its share rises from 20% to 46%.
Put differently, the 3-to-4-word block has become the centre of gravity of search: close to one impression in two, and close to one conversion in two. Nineteen months earlier, that same block was a minority on both counts.
Why queries are getting longer
Tabeling links the shift to the spread of conversational interfaces: large language model assistants, Gemini, and the rollout of AI Mode in results. The hypothesis fits observed behaviour. Once you get used to writing a full sentence to a machine that answers in sentences, you stop writing in telegraph style in every other search box.
That said, it pays to be precise about what the analysis proves. It establishes a change in distribution, not a causal link: the dataset does not separately quantify how much is attributable to AI Mode, to conversational assistants, or to users simply getting more comfortable. The shift sits alongside other signals we have documented, notably the way AI Mode is already redistributing the results surface.
Do you know which phrasings you are actually visible on ?
We analyse your real queries, your presence in AI answers, and the phrasings you do not cover yet.
Why this matters even if you never run ads
The temptation is to file this study under paid search. That would miss the point: Google Ads query data is one of the few public windows into how people phrase their searches, and it comes from the same search box as organic traffic. Paid is measuring a behaviour here, not a channel.
If your keyword plan was built on one- and two-word head terms, you are optimising for a shrinking surface. And the stakes double in generative search: a 3-to-4-word query carries explicit intent, which means a question a page can answer word for word. That is exactly the material answer engines cite, as the dominance of question formats in AI-generated answers shows.
What to do this week
Four concrete actions, in order.
- Measure the shift on your own data. Search Console, Performance report, Queries tab, export over 16 months. Segment by word count and compare the 2025 and 2026 distributions. Twenty minutes is enough, and you will know whether your audience moved as fast as the study sample.
- Reopen your single-keyword pages. The ones targeting one generic term. Add the 3-to-4-word phrasings your users actually type, in the H2s and as direct two-to-three-sentence answers.
- Stop arbitrating on volume alone. A short high-volume term is losing both surface and conversions. A longer, less-searched phrasing captures more of both. Volume has become a poor sole judge.
- Check that every page answers a formulated question, not just that it covers a topic. This is the pivot that serves classic ranking and AI citation at once, and the backbone of our GEO framework.
The limits of what we know
This analysis deserves to be read with its caveats, and they are real. The article publishes neither the sample size, nor the accounts analysed, nor the verticals or geographies covered: there is no way to verify how representative it is. This is paid search data; nothing guarantees the organic distribution moves at exactly the same pace. The link to AI Mode remains the author's hypothesis, not a demonstration.
One last, less comfortable point: secondary coverage of this study does not agree on the conversion share attributed to 1-to-2-word queries, with two incompatible sets of figures circulating side by side. We cross-checked every figure in this article against two independent write-ups before publishing it, and we cannot verify the ones that disagree: we cite only the values that are consistent across both, and discard the rest. Finally, the sample is likely English-speaking and no single national market is broken out: the trend is plausible for you, its speed is not automatically so.
The Cicero take
The short keyword was a convenience of the trade: quick to write, easy to file in a spreadsheet, comfortable to present in a meeting. It was never how people think about their problem. What this study measures is not a phrasing fashion, it is the end of an approximation that SEO tooling has imposed for twenty years. Users finally get to ask their question in full, and they are doing it. The pages that win will not be the ones adding words to their tags: they will be the ones answering a precise question better than anyone else, and cleanly enough that a machine can cite them without rewriting.
Sources
- → Search Engine Land, September 11, 2026: primary source. Analysis by Jason Tabeling (Head of Solutions, Further) of Google Ads query data, January 2025 to August 2026. Impression share of 1-2-word queries from 42% to 24% (18 points in absolute terms); 3-4-word queries from 33% to 48% in impressions and from 20% to 46% in conversions. Sample size and methodology are not published.
- → Storyboard18, September 2026: independent coverage of the same dataset, used here to cross-check the impression shares (1-2 words 42% → 24%, 3-4 words 33% → 48%) and the conversion share of 3-4-word queries (20% → 46%).
Frequently asked questions
Have short keywords become useless?
No. They still account for 24% of impressions according to the analysis published on September 11, 2026 on Search Engine Land. They are simply no longer the centre of gravity, which is a different thing. The right reading is not to abandon them but to stop concentrating most of the effort there: the 3-to-4-word block now captures 48% of impressions and 46% of conversions.
Does Google Ads data apply to organic search?
Partially, and that should be said honestly. The data comes from paid search, and nothing guarantees the organic distribution moves at exactly the same pace. What it does measure, however, is typing behaviour in the same search box that generates your organic traffic. The signal is therefore relevant; transposing the exact figures is not automatically safe: check the distribution on your own Search Console data.
Is AI Mode responsible for longer queries?
The author of the analysis links the movement to the adoption of conversational interfaces, including AI Mode and Gemini, but the dataset does not isolate the contribution of each factor. It establishes a change in distribution, not causation. It is plausible that the habit of writing full sentences to an assistant transfers to classic search; it is not demonstrated by these figures.
How do I check the trend on my own site?
In Search Console, open the Performance report, Queries tab, and export the data over 16 months. Then segment queries by word count and compare the split between 2025 and 2026. The exercise takes about twenty minutes and gives an answer specific to your audience, which beats a market figure: it is your distribution, not that of a sample whose composition was never published.
Growth and SEO & GEO content strategist, I founded Cicéro to help businesses build lasting organic visibility, on Google and in AI-generated answers alike. Every piece of content we produce is designed to convert, not just to exist.
LinkedIn