syted

syted blog

How Long Should a Blog Post Be? The SEO Data, Measured

Every guide ranking for "how long should a blog post be" hands you a table of word counts. One says 2,450 and up. One says 1,500 to 2,000. One says 2,500 minimum in every niche except tech.

None of the three is longer than 1,700 words.

That is not a cheap shot. It is the first measurement in this article, and it points at the thing the whole category gets backwards. Length is not an input you choose before writing. It is an output of how many distinct questions one page has to answer, and that number is sitting in your keyword data before you write a word.

Everything below was measured on 2026-11-22. Keyword and SERP figures come from Ahrefs with country: us. Google quotes come from the live documentation, with the last-updated date each page shows. The word counts of competing pages are my own estimates of body text, and I say so each time.

The guides that rank recommend more words than they contain

Here is the top of the SERP for the exact query, with my estimate of each page's body word count next to the length it tells you to write.

Page Position Est. body words What it recommends
thewritepractice.com 4 ~1,700 "2,450 or more" for Google traffic
wordstream.com 5 not retrievable "3 rules" (page returned HTTP 403)
11content.com 7 ~1,500 2,500 minimum, every niche except tech
mailchimp.com 9 ~1,500 1,500 to 2,000 for a general post

Three of the four pages I could read recommend a length they do not meet. The gap is not small either: the page at position 4 is roughly 700 words short of its own floor.

The honest reading is not "they are hypocrites." It is that their own ranking disproves their advice, on the one query where they would most want it to hold. If 2,450 words were the price of entry, a 1,700-word page would not be the highest organic result.

One more thing about that table. The 11content page carries "In 2024" in its title and a "Last updated 11 Sep 2026" line in its body, and it still ranks at 7. Freshness signals and length advice are being sold by a page that has not reconciled either.

Google's documentation calls writing to a word count a red flag

This is not an interpretation of a conference talk or a tweet. It is a line in the self-assessment list on Google's "Creating helpful, reliable, people-first content" page, last updated 2026-10-05, in the section titled "Avoid creating search engine-first content":

Are you writing to a particular word count because you've heard or read that Google has a preferred word count? (No, we don't.)

Read where that sentence sits. It is not in a tips section. It is in the list of behaviours Google describes as writing for the search engine rather than the reader, next to "producing lots of content on many different topics in hopes that some of it might perform well."

So the advice in the SERP above is not merely unsupported. The exact behaviour it asks for is enumerated by Google as a symptom of the problem. That is an unusually clean contradiction, and it is free to verify.

Two practical consequences follow. A word-count floor in your brief is a liability, not a safeguard. And the "write 2,000+ words to compete" instruction that most content strategy templates still carry is the single line worth deleting from yours.

The page winning this query is a Reddit thread with four referring domains

The organic results for "how long should a blog post be", with Ahrefs' estimated monthly organic traffic for the whole page:

Pos Result DR Ref. domains Est. traffic
2 reddit.com/r/SEO thread 95 4 660
4 thewritepractice.com 72 104 723
5 wordstream.com 90 24 160
6 medium.com 94 6 77
7 11content.com 52 7 102
8 linkedin.com post 99 4 47
9 mailchimp.com 93 5 158
10 youtube.com 99 n/a 33

A forum thread with five backlinks and four referring domains holds position 2. The five polished guides below it (wordstream, medium, 11content, linkedin, mailchimp) add up to 544 estimated visits between them. The Reddit thread alone is credited with 660.

The structural reserve, which the guides never write: Ahrefs' traffic column estimates what the whole page earns across every keyword it ranks for, not clicks from this one query. It proves position plus plausible attribution, not clicks.

That reserve is unusually weak here, and it is worth saying why. Every single organic result on this SERP carries the same top_keyword: "how long should a blog post be", volume 700. No page is here by accident from a larger query. In the four SERPs we measured over the previous fortnight, that was never true.

So the comparison stands. On a query about how much to write, the winning answers are a thread of people typing two paragraphs each and a 1,700-word essay by a fiction writer.

Length is an output. The input is the cluster

Here is the number that actually decides how long a post should be, and almost nobody uses it.

Ahrefs' traffic potential estimates what the current number-one page earns from every keyword it ranks for. Compare it to the keyword's own volume and you learn something the keyword alone cannot tell you: whether winning this query also wins a cluster, or just this query.

Eleven keywords, measured the same day, same country:

Keyword Volume KD Traffic potential TP / volume
canonical tag seo 606 6 6,000 9.9
what is a 301 redirect 823 67 3,300 4.0
google business profile optimization 4,266 35 13,000 3.0
how to do keyword research 3,189 77 9,200 2.9
what is alt text 2,199 11 5,000 2.3
how long does seo take 641 8 1,100 1.7
what is schema markup 1,615 68 2,600 1.6
meta description length 2,585 26 4,000 1.5
how long should a blog post be 578 7 600 1.04
seo for plumbers 2,117 5 600 0.28
ecommerce seo checklist 577 11 150 0.26

The ratio spans a factor of 38 across queries that look alike in a spreadsheet. Three bands fall out of it, and each one implies a different length.

Ratio well above 2. The page that wins this query also wins a crowd of neighbours. canonical tag seo has volume 606 and a parent topic of rel=canonical worth 2,500. To hold that position you have to answer cross-domain canonicals, self-referencing canonicals, canonical versus noindex, and canonical versus 301. Length is the consequence of that list, not a target you set.

Ratio near 1. Winning this query wins roughly this query. The parent topic of "how long should a blog post be" is itself, at 700. There is no cluster behind it to capture. A 3,000-word page here buys nothing that a direct answer plus the evidence does not already buy.

Ratio below 1. The demand exists but it does not land on an article. seo for plumbers has 2,117 searches and a traffic potential of 600, and the reason is in the SERP rather than the keyword: the map pack takes the clicks. We wrote the plumbing version of that finding in full, and the conclusion generalises across local trades.

How to read the ratio before you write

The method is four steps and it takes about five minutes per keyword.

  1. Pull volume, difficulty, traffic potential and parent topic for the target keyword.
  2. Divide traffic potential by volume.
  3. Open the parent topic. If it is a different, larger keyword, that keyword is the real page you are being asked to write.
  4. Count the distinct questions the parent groups. That count, not a word target, sets the outline.

Step 3 catches an expensive mistake. what is alt text has a parent of alt text examples at 2,800, which means the page that wins is an examples page, not a definition page. what is a 301 redirect has a parent of http 301. In both cases the searcher's question and the winning page's format are not the same thing, and writing more words about the definition will not close that gap.

A keyword whose parent is much larger than itself says the demand lives on another page. A keyword whose parent is much smaller says the demand does not live on a page at all. We hit the second case in full on a nonprofit query where the parent was a thousand times smaller than the keyword, and the answer turned out to be the map pack again.

Outline first, from the cluster. Word count never. If you need the outline to be a repeatable artefact rather than a habit, our content strategy template is built around exactly this step.

Measure the SERP instead of trusting a table

Every length table in this category is an average of somebody else's SERP. Yours is one fetch away. This is dependency-free, so it runs on any machine with Node 18 or newer.

// serp-lengths.mjs
// usage: node serp-lengths.mjs urls.txt
import { readFileSync } from 'node:fs'

const urls = readFileSync(process.argv[2], 'utf8')
  .split('\n').map(s => s.trim()).filter(Boolean)

function bodyWords(html) {
  const scoped = html.match(/<(article|main)\b[^>]*>([\s\S]*?)<\/\1>/i)
  const source = scoped ? scoped[2] : html
  return source
    .replace(/<(script|style|nav|header|footer|aside|form|noscript)\b[\s\S]*?<\/\1>/gi, ' ')
    .replace(/<[^>]+>/g, ' ')
    .replace(/&[a-z#0-9]+;/gi, ' ')
    .split(/\s+/)
    .filter(w => /[a-z0-9]/i.test(w))
    .length
}

for (const url of urls) {
  try {
    const res = await fetch(url, {
      redirect: 'follow',
      headers: { 'user-agent': 'Mozilla/5.0 (compatible; length-audit/1.0)' },
    })
    if (!res.ok) { console.log(`HTTP ${res.status}\t${url}`); continue }
    console.log(`${bodyWords(await res.text())}\t${url}`)
  } catch (err) {
    console.log(`ERROR\t${url}\t${err.message}`)
  }
}

Three honest limits, all of which I hit while writing this article. Stripping HTML with regular expressions is approximate, and a page without <article> or <main> will carry some navigation into the count. Some sites refuse the request: wordstream.com returned HTTP 403 today, which is why its row above is blank. And a number you got from one fetch is a snapshot, not a trend.

Even so, this beats the tables. Run it on the ten URLs currently ranking for your keyword and you get the distribution that matters, which is the one Google is actually rewarding on your query this week.

What you are looking for is not the average. It is the spread. A SERP where the lengths cluster tightly is telling you the format is settled. A SERP that runs from 400 words to 4,000, which is what this one does, is telling you length is not what separates the winners.

Check afterwards whether the length bought the cluster

The test of a long post is not whether it ranks. It is whether it earns impressions on more than the one query you targeted. If your 3,000-word guide shows up for four queries, the extra 2,000 words bought you nothing and you should know that.

Search Console answers it directly. One request, filtered to the page, grouped by query:

SITE="sc-domain:example.com"
PAGE="https://example.com/blog/your-post"
END=$(date -u +%F)
START=$(date -u -d '90 days ago' +%F)

curl -s -X POST \
  "https://www.googleapis.com/webmasters/v3/sites/$(jq -rn --arg s "$SITE" '$s|@uri')/searchAnalytics/query" \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  -d "$(jq -n --arg start "$START" --arg end "$END" --arg page "$PAGE" '{
        startDate: $start,
        endDate: $end,
        dimensions: ["query"],
        rowLimit: 25000,
        dimensionFilterGroups: [
          { filters: [{ dimension: "page", operator: "equals", expression: $page }] }
        ]
      }')" | jq '.rows | length'

That integer is your answer. Compare it against the parent topic you used to justify the length in the first place. A page written to capture a parent worth 2,500 searches and returning nine queries did not capture it.

Two caveats before you read too much into the number. Google's own API reference states the endpoint "does not guarantee to return all data rows but rather top ones", so treat the count as a floor. And rowLimit tops out at 25,000, which you will not come close to on a single page.

Run this across your archive and the pattern usually separates cleanly: a few pages ranking for hundreds of queries, a long tail ranking for three or four. The second group is where length was spent and not recovered. That is the audit, and it is more useful than any rewrite. If none of your pages return anything at all, the problem is upstream of length, and the indexing checklist is the place to start.

What the AI Overview changes, and what it does not

This query carries an AI Overview at position 1, with nine sitelinks under it, on 2026-11-22. The summary answers "how long should a blog post be" before any of the eight organic results gets a chance.

The temptation is to conclude that longer posts give the summary more to chew on. Google's own documentation does not support it. The "AI features and your website" page, last updated 2025-12-10, says plainly:

There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.

The same page adds that eligibility as a supporting link requires only that a page be "indexed and eligible to be shown in Google Search with a snippet." No length threshold appears anywhere on it.

What the AI Overview does change is the economics of the band near ratio 1. When a query is self-contained and a summary answers it above the fold, the marginal value of your 2,000th word is close to zero, because the click it was supposed to earn is being absorbed before the result list. The same logic applies to answer engine optimization more broadly: the unit that gets quoted is the passage, not the page.

Here is the uncomfortable half. The thing that wins both a summary citation and a human reader is a specific, verifiable claim with its source attached, and claims like that are expensive to produce. A long post made of general advice is cheap and useless. A short post made of measurements is expensive and useful. Length is the wrong axis entirely, and no one can guarantee that getting it right produces a position either way.

What this means for your next brief

Replace the word-count line in your brief with three lines.

  • Parent topic and its volume. This is the real scope of the page.
  • Questions the parent groups. Pulled from People Also Ask and from the subheadings of the pages ranking now. This becomes the outline.
  • The evidence each section will carry. A number, a quote with a date, a screenshot, a command. Sections that cannot name their evidence get cut, which is where most of the padding disappears.

Written this way, a post lands where it lands. The article you are reading came out around 3,000 words because the measurements took that much room, and the query sits at ratio 1.04, so it would have been wrong to stretch it to 4,000.

The reverse case is just as common and less obvious. A post that turns out to be 700 words because the question has one answer is not thin, whatever an audit tool scores it. Thin means missing the thing the searcher came for. A 2,500-word page that never answers the question is the thin one, and it is the one most SEO content writing processes reliably produce.

This is also why publishing cadence and length are the same argument in different clothes. Both treat a quantity you control as a proxy for a quality you do not. Both survive because the proxy is easy to measure and the quality is not. Neither is a ranking signal, and how long search takes to work does not change because you added paragraphs.

Set the scope from the cluster, write until the evidence is covered, stop. Then measure what the page actually earned, with the Search Console call above, and let that decide the next one. If you want that loop written down as a repeatable process rather than a principle, we published the whole workflow separately.

FAQ

What is the ideal length for a blog post?

There is not one, and Google's documentation says so in a line aimed directly at the question: "Are you writing to a particular word count because you've heard or read that Google has a preferred word count? (No, we don't.)" The useful substitute is the parent topic of your keyword. If the parent groups twelve related queries, the page has twelve things to answer and the length follows. If the parent is the keyword itself, as it is for this one, answering it well is the whole job.

Is a 3,000 word blog post too long?

It is too long if it is 3,000 words about a question with a 600-word answer, and about right if the parent topic it targets groups a dozen queries. The test is not the count, it is the ratio of traffic potential to volume and the number of distinct questions behind it. Run the Search Console query above on an existing 3,000-word page: if it earns impressions on five queries, the length did not buy what it was supposed to buy.

Does word count affect Google rankings?

Not as a direct input. The correlation studies that get quoted everywhere, including the "2,450 words" figure sitting at position 4 of this SERP, measure that pages covering a subject thoroughly tend to be longer. They do not show that adding words causes rankings. The SERP measured here is evidence against the causal reading: a Reddit thread with four referring domains outranks three guides of 1,500 words or more.

How many words should a blog post be for SEO?

Enough to answer every question the parent topic groups, and not one more. In practice that means deciding the outline first from the cluster, then writing to the outline. A post whose length was decided before its outline is the exact pattern Google's helpful content guidance flags as search-engine-first. Nobody can guarantee a position, so the only defensible target is coverage of the question.

What is the 80/20 rule for blogging?

In its usual form it claims 80 percent of your results come from 20 percent of your posts, and in our measurements that split is real but is about query coverage rather than length. A handful of pages pick up hundreds of queries each and the rest pick up a few. The 20 percent are almost never the longest posts. They are the ones whose parent topic had a large cluster behind it, which you can check before writing rather than discovering a year later during a content audit.

Are blogs still worth it in 2026?

They are worth it when the page answers a question that a cluster of people are actually asking, and are not when the page exists to hit a publishing quota. The arrival of AI Overviews, present on this very query, shifts the value toward specific and verifiable claims and away from general guidance, because general guidance is what the summary already does. That is a change in what to write, not an argument for writing more or longer.

Get cited by ChatGPT. Rank on Google.

You found this article through search. That is the whole product.

  • One researched article a day
  • Published on your own domain
  • Keywords checked against live results
Start writing

Get cited by ChatGPT. Rank on Google.

You found this article through search. That is the whole product.

  • One researched article a day
  • Published on your own domain
  • Keywords checked against live results
Start writing