Key takeaways
- Check what renders before you check anything else. If the body only exists after JavaScript runs, every other analysis is of a page nothing can read.
- Then check extractability: can one passage answer the page main question on its own, without the surrounding context?
- Then check attribution. The GEO research found citations, quotations and statistics lifted source visibility by up to 40% in its test harness, which makes attribution a format decision with measured effect.
- Then check what the page is unambiguously about. A page covering four topics competently is a worse answer to each than four pages would be.
- Keyword counts come last and matter least. Google own guidance is about helpful, people-first content, and a density target is the opposite of that.
- A tool gives you the counts. Deciding what a count means is the judgement, and that is the part no tool supplies.
Content analysis has become a list of numbers: word count, keyword density, readability grade. Those are measurable, which is why they are measured, and they are not what decides whether a page gets used.
What decides it is whether the page can be read at all, whether an answer can be lifted out of it, and whether its claims are traceable. Those are harder to score and they are the actual job.
This is the order we check a page in, and why that order rather than another.
What should you check first on a page?
Short answer: What renders without JavaScript. Load the page with scripts disabled and look at what remains, because if the body content only appears after a script runs then everything downstream of that is analysis of a page nothing can actually read.
This takes thirty seconds and invalidates most content audits that skip it. A page can look complete to every human reviewer and be empty to a crawler.
Google documentation states that its AI features draw on core ranking systems, which means an unreadable page is unreadable to both surfaces at once. There is no separate path in for content a crawler cannot see.
We check this first on every page we look at, including our own, because the failure is invisible and the fix is architectural rather than editorial.
What makes content extractable?
Short answer: A passage that answers the question on its own. If the answer to the page main question is distributed across four paragraphs, something summarising the page has to reconstruct it, and reconstruction is where a brand gets dropped from the summary.
The practical form is a short, self-contained answer near the top of each section, followed by the explanation. Forty to sixty words is the range we use, because it is long enough to answer and short enough to quote.
- Self-contained. It reads correctly with nothing before or after it. Pronouns referring to earlier paragraphs break this.
- Positioned early. Immediately under the heading that asks the question, not at the end of the section.
- Question-shaped heading above it. So the match between question and answer is structural rather than inferred.
- Not the whole section. The answer is the summary; the section still has to earn the reader attention afterwards.
The first item is where most attempts fail. An answer beginning with the word this is an answer that cannot be lifted.
Why does attribution change anything?
Short answer: Because it is one of the few format decisions with measured effect. The GEO paper found that adding citations, quotations and statistics raised a source visibility by up to 40% in their test harness, which is a research result on a specific setup rather than a universal law.
The mechanism is plausible even without the measurement. A passage with a named source attached is safer to repeat than one that asserts something anonymously.
In practice this means naming your sources inline rather than in a footer list. A claim attributed where it appears travels with the attribution when it is quoted.
It also means being explicit when something is your own observation rather than a cited fact. We do that throughout our own writing, and it is the cheapest credibility signal available.
How do you tell what a page is about?
Short answer: Read the title, the first heading and the first fifty words, then say in one sentence what the page answers. If you cannot, the page covers several things, and a page covering four topics competently is a worse answer to each than four pages would be.
This is the diagnosis behind most underperforming content. The page is not bad; it is three pages that were never separated.
The symptom in your own data is impressions spread thinly across unrelated queries with no position on any of them. That is a page being considered for several things and chosen for none.
Splitting such a page usually beats improving it. Each half can then be unambiguous, and unambiguous is what both ranking and extraction reward.
Where do keyword counts fit?
Short answer: Last, and as a sanity check rather than a target. A page about a subject will naturally contain the words for that subject, so a low count usually means the page is vague rather than that it needs words inserted.
The useful version of the check is absence rather than density. If a page about a service never uses the phrase customers use for that service, that is worth fixing.
The useless version is a target. Google guidance is about helpful, people-first content, and writing to hit a density figure produces the opposite.
Snippets are a related trap. Google documentation says it generates snippets from page content and may ignore your meta description, so writing the description as a keyword container achieves nothing.
Frequently asked questions
How long should a page be?
As long as the subject needs. Length correlates with performance because thorough pages tend to be longer, not because length causes anything. A padded page is worse than a short one, because padding dilutes what the page is about.
Does readability score matter?
Less than clarity, and the two are not the same. A technical page written for specialists will score badly and be correct for its audience. Write for the reader you need, then make the structure help everyone else skip to their part.
Should every section have a short answer block?
Every section that answers a question, which should be most of them. A section that is purely explanatory does not need one, but if you cannot write a self-contained answer for a section it may not have a clear point.
Can a tool tell me if my content is extractable?
It can tell you whether the structure exists: headings, answer blocks, paragraph lengths. Whether the answer is actually self-contained requires reading it, which is judgement rather than measurement.
What is the single most common content problem?
A page that covers several topics instead of one. It is also the most expensive to fix, because the remedy is splitting rather than editing, and splitting means deciding which URL keeps the history.
Sources and further reading
- Google Search Central, AI features and your website
- Google Search Central, creating helpful, reliable, people-first content
- Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024
- Google Search Central, how to control your snippets in search results
- Google Search Central, getting started with Search Console
Run a page through the checks
Our content analysis tool gives you the counts for free. The five checks above are what to do with them, and we will walk a page through them with you if the counts are not telling you enough.