Word count is a floor to clear, not a target to hit
There is no word count that ranks. Every number you have been given, 300, 800, 1,500, 2,400, was measured off pages that already ranked, which tells you what winning pages happened to weigh and nothing about why they won.
The short answer
There is no ideal word count, and every number published as one comes from a correlation study rather than from anything an engine measures. What a low count is genuine evidence of is a page that may not cover its subject, which is a different claim. Judge a page against what the query needs, not against an average.
Where the recommended numbers come from
The studies are real and the inference from them is not. Take the top ten results for a few thousand queries, count the words, average them. The output is a number, and the number is reported as a target. What was actually measured is the length of pages that satisfied competitive queries, on subjects that happened to be in the sample, in a year that has passed.
The confusion is a correlation read backwards. Comprehensive treatments of complicated subjects tend to be long, and they tend to rank, so length and ranking move together. Adding words to a page does not make it comprehensive, in the same way that a taller thermometer does not make a room warmer.
You can see the failure directly in the queries where the winning page is short. A conversion question, a definition, a syntax reminder, an opening-hours lookup: the results are two hundred words each, because two hundred words is the whole answer. A 2,000-word page on any of those loses to the short one, and no amount of thoroughness fixes a page that made someone scroll for a sentence.
The one sentence to take from this section: match the length of the answer, not the length of the competition.
If the question needs eighty words, write eighty. If it needs four thousand because there are six cases and each has an exception, write four thousand and use headings so nobody has to read the parts that do not apply to them. Both of those are correct, and the number in between is not a target either of them was aiming at.
What a low count is genuine evidence of
Dismissing the metric entirely goes too far. A count is weak evidence of quality and decent evidence of a few specific things, all of them mechanical.
- Below about 150 words, the page has not established its subject. Not as a rule about ranking, as a practical matter: there is not enough text for anything to determine what the page is for, including a reader who landed on it by accident.
- A count near zero on a page you know has text means the text is not in the HTML. This is the most useful thing the metric does. A visually complete page reporting forty words is a page whose content arrives by script, and the finding you actually have is a rendering problem.
- A whole template clustered at one low number is a template, not a decision. Four hundred pages all between 90 and 120 words were generated. Nobody wrote four hundred pages that happened to land in a thirty-word band.
- A high count on a page with no headings is a readability finding wearing a length badge. Two thousand words in eleven paragraphs with no structure is not a length problem, and reporting it as one hides what is wrong.
Notice that three of the four are about finding a different problem. That is the honest value of this check: it is a cheap signal that points elsewhere, and it is nearly worthless as a verdict on its own.
Count the words on one page, then interpret them
This reports the count for the URL you enter and puts it in a band. The bands are deliberately coarse, because a precise threshold would be a claim this guide has just spent two sections arguing against.
No signup required. Each free search audits one page, paste any URL to see it in action.
Read the number as a prompt rather than a grade. A low count asks whether the text is really there and whether the page has a subject. A high count asks nothing at all.
Your navigation is in that number
A word count taken from a rendered page counts every word in it: the menu, the sidebar, the promotional block, the cookie notice, the four columns of footer links and the legal paragraph. On a heavy template that is 300 to 600 words before your content starts.
The consequence is that the metric is inflated in exactly the situation where you most need it to be accurate. A generated page with one unique paragraph reports 480 words, clears every threshold anyone recommends, and is empty. That gap is what the uniqueness ratio exists to measure, and it is why a length check on its own can never find a thin page.
Two practical adjustments. When you compare pages, compare pages sharing a template, so the boilerplate cancels out and the difference between the numbers is real. And when a number matters enough to act on, take it from the article region rather than the document, which is the last section of this guide.
Does this affect AI search?
Length has no direct bearing on retrieval, and the mechanism is worth knowing because it changes the advice. Retrieval works on chunks: a document is split into passages and the passages are matched against a question. So the unit competing for your answer is a few hundred words, not the page.
That has two consequences pointing in opposite directions. A very short page produces one thin chunk with little around it to establish context, which is a weak candidate. A very long page produces many chunks, most of which are irrelevant to any given question, and being long buys you nothing except more chances to be irrelevant. What wins in both cases is a self-contained passage that resolves a question, which is a property of paragraphs rather than of pages.
Which turns the usual instinct inside out. If you want to be quoted, the useful work is not adding a thousand words. It is making sure the four hundred you have stand on their own without the page around them. A long page that only makes sense read in order is harder to quote than a short one that answers in the first paragraph.
Why a metric that proves nothing is still rated Important
Because of what it catches at the bottom of the range. A page reporting sixty words is either empty or broken, and both of those are real problems that this check finds instantly and cheaply. Rating it Important is a statement about the low end, not an endorsement of word counts as a quality measure.
It stays below Critical because a short page is not excluded from anything. It is crawled, indexed and served, and if it answers a narrow question it can outrank everything else on the page. There is no mechanism here that removes a URL from the results, which is where the top of the scale is reserved for.
The rating is also a compromise made necessary by how the finding gets misused. Rated any higher, it would encourage padding, which actively damages pages. Rated any lower, sites would keep shipping templates that emit ninety words and never notice. Important, with the guide attached explaining that the number is not a target, is the position that survives both failure modes.
Counting only the words that are yours
Two counts, and the difference between them is the number that means something. If your template marks its content region, this takes about a minute.
URL=https://example.com/guides/sharpening
text() { sed -e 's/<[^>]*>/ /g' | tr -s ' \n' ' '; }
# the whole document, which is the number most tools report
whole=$(curl -s "$URL" | text | wc -w)
# just the article, if the template marks one
body=$(curl -s "$URL" \
| tr '\n' ' ' \
| grep -o '<article.*</article>' \
| text | wc -w)
echo "document: $whole words"
echo "article : $body words"
echo "furniture: $(( whole - body )) words you did not write for this page"If the second number comes back as zero, your template has no <article> or <main> to find, which is worth fixing for its own sake: the same element that makes this measurable is the one that lets a reader mode, a chunker and an assistive technology find your content.
Compare the furniture number against the article number across a few templates. Where the furniture wins, you have learned something the word count on its own was actively hiding.
Questions this check raises
- How many words should a page have?
- As many as the subject needs and no more. The 1,500 and 2,000 word figures circulating come from studies correlating length with ranking on competitive queries, where long pages win because those subjects need length, not because length wins. A definition page answering one question in 200 words is complete.
- Is thin content the same as short content?
- No, and conflating them causes real damage. Short means few words; thin means the page does not deliver what it promises. A 100-word answer that fully resolves the query is short and good. A 900-word page that circles the subject without answering it is long and thin.
- Does my navigation count toward the word count?
- It should not, and a naive count includes it. A site with a large menu and a long footer can show 400 words on a page whose actual content is 60, which makes every count on the site look healthier than it is. Counting only the words inside the main content region is what makes the number mean anything.