TL;DR
- Use enough detail to answer the buyer's question, support the answer and explain its limits. A fixed word target cannot do that job.
- Cited-page length and citation position do not measure a page's chance of being selected in the first place.
- Research on where cited passages appear supports a clear opening, but it does not make the first 100 words a universal cutoff.
- Add missing evidence and useful detail; remove repetition. Check the answer before counting the words.
Start with the missing answer
The better editorial question is what the reader still cannot judge. More words can help when they add a test method, a condition or a meaningful trade-off. They add little when they repeat the same claim in new language.
What cited-page length actually tells us
Ahrefs analyzed 174,048 extractable pages found through Google AI Overview citations. Its reported length bands were 16.6% under 350 words, 36.8% at 350–1,000, 30.6% at 1,000–2,000 and 16% over 2,000. The analysis removed HTML and boilerplate before counting words.
| Reported word-count band | Share of extracted cited pages |
|---|---|
| Under 350 words | 16.6% |
| 350–1,000 words | 36.8% |
| 1,000–2,000 words | 30.6% |
| Over 2,000 words | 16% |
Source: Ahrefs, December 2025. Bands follow the source's labels. This is a selected set of cited pages, not all pages competing for those questions.
Both short and long pages occur in the sample. That is useful evidence against a claim that one format is excluded. It does not show that the most common band gives a new page the highest chance of citation. We would need a suitable comparison with uncited pages to make that judgment.
This is a basic denominator problem. Suppose a market publishes many more short pages than long ones. Short pages could supply more citations even if their selection rate were lower. That example explains the measurement issue; it is not a finding about the actual supply of pages.
Google AI Overview sources that could be extracted. The distribution does not measure selection rates among all eligible pages. Ahrefs, December 2025.
Do not confuse citation position with selection
Ahrefs reports a 0.04 Spearman correlation between word count and citation position within AI Overviews. That result concerns where cited pages appeared in the citation list. It should not be restated as proof that length has no effect on whether a page is selected.
For a marketing team, the distinction changes the content brief. The study gives no sound reason to demand a long article merely to obtain a prominent citation. It also gives no reason to cut necessary detail until the page fits a short template. It leaves the quality and scope of the answer to be assessed.
A clear opening is a different question
Surfer's June 2026 collection mapped 100,000 citation placements from 10,000 AI Overview responses. It reported that 38% of placements were in the source page's first 100 words. This describes where matched passages occurred, rather than the ideal length of the full page.
Surfer labels the evidence correlational. Its explanation that earlier passages cost less to retrieve is a proposed mechanism, not a demonstrated rule of every AI system. The finding does support checking whether your opening gives readers a direct answer instead of making them work through a long preamble.
| Evidence | Result and measured object | What it helps assess | What remains untested |
|---|---|---|---|
| Ahrefs, December 2025 | 16.6% of extracted cited pages under 350 words; word count vs. citation position correlation 0.04 | The range of page lengths present in citations and their ordering | Selection rates among all eligible pages |
| Surfer, September 2026 update | 38% of matched citation placements in the first 100 words; June 2026 collection | Where to look for buried answers | A causal gain from moving text, or an ideal total length |
The studies complement each other without establishing a winning word count. One counts pages and their citation order; the other maps passages. Both vendors sell content or search tools, and both use their own collection methods. Neither result is a B2B-wide rule for every platform.
Build the brief around the decision
Use this editorial scope table to decide what earns space on a page. Each row needs a concrete answer before a word target becomes useful.
| Scope question | Evidence to include | Reason to add detail | Reason to stop |
|---|---|---|---|
| What is the buyer deciding? | One explicit decision or question | Readers cannot tell whether the answer applies | The main answer is clear |
| Under what conditions is it true? | Inputs, versions, scope or exclusions | An omitted condition could change the decision | Material limits are stated |
| Why should the buyer believe it? | A relevant test, source or worked example | The claim cannot be checked | Evidence supports the actual claim |
| What must the reader do next? | A useful check or appropriate contact path | The answer leaves a necessary task unresolved | The next action is clear and relevant |
Imagine a fictional software supplier explaining whether an integration supports a buyer's deployment. The draft has a long history of the product but omits the supported version and an important limitation. Adding another generic paragraph will not fix it.
A useful revision puts the direct compatibility answer near the start, adds the version and limit, and links to the approved setup guide. It removes history that does not help the decision. The result might become shorter or longer. Its value comes from what the buyer can now judge, not the change in length.
Content Length and AI Search: Common Questions
Is 1,000 words a useful target for every B2B article?
No. It can be a planning estimate for a defined assignment, but it should not decide which evidence stays or goes. Google says it has no preferred word count. That guidance applies to Google Search, not every AI product.
Should the whole answer fit in the opening paragraph?
Give the direct answer and its main condition early. Detailed proof, exceptions and examples can follow. A useful opening helps a reader decide to keep reading; it does not replace the rest of a complex explanation.
Can we split a long guide into many short pages?
Split it when the sections serve distinct reader tasks and can stand alone. Keep related evidence together when separating it would remove context or repeat the same answer. Page count is not a substitute for useful coverage.
How should we test a length change?
Define the intended improvement first, such as clearer scope or a missing answer. Track citations under consistent questions and conditions, alongside buyer use. If scope, layout and evidence all change, report the combined revision rather than crediting the word count alone.
Review the scope of your page
If your briefs rely on word quotas while buyers still lack clear answers, bring a page and the decision it should support to SemShift. We can discuss gaps in evidence and scope before you invest in expanding the content library.