People Are Typing Longer Searches. Your Keyword List Probably Has Not Noticed.
A dataset published in Search Engine Land in September 2026 says short searches are dying and nobody told your keyword list.
Between January 2025 and August 2026, one- and two-word queries fell from 42% of all ad impressions in that dataset to 24% — from nearly half of everything the ads appeared on to about a quarter. Three- and four-word queries went the other way, 33% up to 48%. On conversions the move is bigger: three- and four-word queries went from 20% of conversions to 46%, while short queries slid from 62% to 52%.
Before you act on any of that, two things need saying. The first is that this is one contributor's dataset, published without a disclosed sample size, account mix, or method — a signal, not a benchmark. The second is bigger: the report you would use to check any of this against your own account changed this year, and for a growing share of searches it is no longer showing you what people typed. That is the part almost nobody connected to this story, and it is why I wrote this.
Would you rather have someone look at your actual search terms than read about somebody else's? Book a free strategy call. Otherwise, keep going — everything here is a report you already have access to.
What the Data Actually Says
Two measures moved, and they did not move by the same amount. Worth separating, because people keep collapsing them into one headline.
| Query length | Share of ad impressions | Share of conversions |
|---|---|---|
| 1–2 words | 42% → 24% | 62% → 52% |
| 3–4 words | 33% → 48% | 20% → 46% |
| 5–6 words | — | 3% → 9% |
| 7+ words | — | 1% → 4% |
A note on that table, because you will notice it before I do. The conversion column does not sum to 100% in either period — 86% at the start, 111% at the end. The source publishes these as four separate figures rather than slices of one pie, and I have left them as published instead of normalizing them into something tidier than the data is. Read each row against itself: where that band was, where it is now.
There is a second wrinkle in the same direction. The article says the conversion share for one- and two-word queries "has collapsed, suffering a -35% absolute change," and then, a few lines later, that the same bucket went "from 62% of conversions" to "just 52% — a 10-point drop." Those are two different claims about one number. I have used the 62-to-52 figures here because they are the specific ones, and I am flagging the contradiction rather than picking the scarier number and moving on.
With that said, read the two columns against each other and the mismatch is the interesting part. Short queries lost roughly 43% of their share of impressions and about a sixth of their share of conversions. They are showing up far less often and still carry the largest single share of conversions on the list.
That is not "short keywords are dead." That is short keywords shrinking as a slice while longer queries grow into real volume. The practical version: the searches that are growing are the ones that sound like sentences. Not "plumber" but "emergency plumber open sunday near me."
The Part That Should Worry You More
Here is the thing almost nobody connected to this story, and it is the reason I wrote the article.
The search terms report is where you would go to check any of this. It is also, for a growing share of searches, no longer showing you what people typed.
Google's own ad group prioritization documentation now says that for certain search types, "the search term shown in your reporting represents the best approximation of the user's intent." PPC Land reported the change in May 2026. It applies to AI Mode queries, AI Overviews, Google Lens searches, and autocomplete selections.
Sit with that for a second. A search term in your report may be Google's summary of what it believes the person meant, rendered as a phrase that person may never have typed.
Which means a chart showing queries getting longer is partly a chart of Google writing longer approximations. I am not saying the whole trend is an artifact — it almost certainly is not, because people really are talking to search engines in sentences now. I am saying the measuring instrument changed during the measurement period, and the article does not mention it.
This matters beyond one chart. If you build negative keyword lists off the search terms report — and you should, it is still the best tool available — you are now blocking approximations as well as actual queries. That is a reason to be a little more generous with what you leave running and a little slower to conclude that a single ugly-looking term is the problem. It is the same discipline as telling a reporting bug from a real traffic drop: know what your report is actually counting before you make an expensive decision about it.
Why You Should Not Take the Numbers at Face Value
Back to the dataset, because I am not going to pretend it is settled. The article does not give you what you would need to know that it is.
There is no stated sample size. No account mix, no industries, no budgets, no countries. No note on whether this is one agency's book of business or something broader. And no attempt to separate correlation from cause — the piece attributes the shift to AI Mode and wider LLM adoption, which is plausible and also exactly what everybody assumes without measuring. Add the column that sums to 111% and the two incompatible figures for the same bucket, and what you have is a directionally interesting chart, not a measurement.
None of that makes the direction wrong. Advertisers have been describing this shift for a year and it is consistent with what I see in accounts. But there is a difference between "this matches what I am seeing" and "this is a benchmark I will restructure an account around," and a lot of marketing advice this month is quietly crossing that line.
The honest position: use this as a reason to go look at your own numbers. Your account is the only dataset whose sample size you can verify.
AI Mode Is Not a Reason to Switch Campaign Types
Short section, because the news here is mostly reassuring.
Google is running a small test that lets exact and phrase match keywords in regular Search campaigns serve text ads inside AI Mode, when the query shows what Google Ads Liaison Ginny Marvin called "explicit and direct user intent." Not just Performance Max. Not just AI Max.
Marvin's own word for it was "experiment." No scale disclosed, no opt-in or opt-out described, no start date given.
So the useful takeaway is a negative one: if someone tells you to move a working Search campaign into Performance Max so you can "be in AI Mode," that is not a reason. A tightly controlled Search campaign may already be eligible. And handing Google more automation costs you real control — the search terms you can act on, the structure you can build around, the ability to say no to a segment. Three platforms already took a piece of your cost controls in recent months. There is no need to volunteer the rest.
What to Do in Your Own Account
Four things, in order. The first three are free — they are you and a spreadsheet. The fourth spends money, so it goes last and only if the first three earn it.
- Pull 90 days of search terms and sort by word count. Export the search terms report, add a column counting words, and pivot on it. Twenty minutes in a spreadsheet. You will know within one screen whether your account looks like that chart or not.
- Compare cost per lead across the bands. Not clicks, not impressions — what a lead cost you in each band. If your three- and four-word terms bring cheaper leads than your two-word terms, that is your own data telling you the same story, and it is worth more than anybody's survey.
- Find the converting long queries you have no ad for. This is where the money usually is. A query like "do i need a permit to replace a water heater" converts because it is a person with a specific problem, and most accounts answer it with an ad written for the word "plumber." Write the ad that answers the question.
- Add phrase match coverage where it is missing. If your account is all exact match, you are unlikely to be catching the longer natural-language variants at all. Phrase match is the middle ground that keeps you in the conversation without handing over the steering wheel. This one costs money, unlike the three above it: phrase match will pick up queries exact match never saw, and some of them will be junk. Add it to one ad group first, watch cost per lead for two weeks, and expand only if the lead cost holds.
Steps one through three are an afternoon of looking at a report you already have, which is the cheapest kind of work there is.
Not sure what your own split looks like?
Export 90 days of search terms, delete any column you do not want to share, and send it to saguarosocialmediaaz@gmail.com. I will tell you how your query lengths break down, whether the longer ones are actually converting for you, and the two or three terms I would write an ad for. One paragraph back, no pitch, and I will not follow up unless you ask me to.
What happens to the file: I read it, I reply, I delete it. It does not go into a CRM, a mailing list, or a case study.
What the export cannot tell me is whether your ad copy and account structure are the reason your longer terms underperform. That takes the account, and that is what a call is for.
What Not to Do
Three tempting moves that will cost you.
Do Not Delete Short Keywords Because They Are Short
Look at the conversion column again. Short queries still carry the largest share of conversions in that dataset. If a two-word keyword brings you leads at a price you are happy with, word count is not a reason to touch it. Judge every keyword on 90 days of your own performance, the same as always.
Do Not Chase Seven-Word Queries
They quadrupled, from 1% of conversions to 4%. Quadrupling something tiny gives you something slightly less tiny. For most small accounts, seven-word queries are a handful of conversions a month and not worth building structure around. The action is in the three-to-four word band, which is where both volume and conversions actually moved.
Do Not Rebuild the Account Around One Chart
Restructuring a working account is expensive in a way that does not show up on an invoice. Every meaningful change puts bid strategies back into learning, which is the period when costs are least predictable. If your account is performing, the right response to this news is a spreadsheet and some new ad copy, not a rebuild. Concentration still beats churn, and that is as true for structure as it is for platforms.
Common Questions About the Query Length Shift
Five worth answering directly.
Should I Delete My Short Keywords?
No. Judge each keyword on what it did in your own account over the last 90 days, not on its word count. A two-word keyword bringing leads at a cost you like is a good keyword. This data is a reason to look, not a reason to delete.
Is This Data Reliable?
Partly. It is one contributor's dataset with no disclosed sample size, account mix, or methodology, its conversion column sums to 111%, and it gives two different figures for the same number. The direction matches what plenty of advertisers report. Treat it as a prompt to check your own account, not as an industry benchmark.
Are the Search Terms in My Report the Actual Words People Typed?
Not always. Google's documentation says that for AI Mode, AI Overviews, Google Lens, and autocomplete selections, the reported search term "represents the best approximation of the user's intent." Some of what you are mining is an inference, not a transcript.
Do I Need Performance Max to Show Up in AI Mode?
No. Google is testing whether exact and phrase match keywords in regular Search campaigns can serve inside AI Mode on high-intent queries. It is an experiment, not a rollout, and not a reason to change campaign types.
What Should I Actually Change This Month?
Pull 90 days of search terms, sort by word count, and look at what each band cost and returned. If longer terms are converting and you have no ad written for them, write those ads. That is the change worth making.