Why this signal matters for Google
Quality is not length, and it is not a clean writing style. It is what the reader finds at your place and nowhere else. Take two pages about a leaking water heater. The first repeats what the top five results say, at greater length. The second shows three photos of worn parts taken on a call-out, the average replacement price in your area, and the sentence no other site writes: past what age it stops being worth repairing. Both pages cover the same topic. Only one brings something. That is where how to create quality content is actually decided: not in the word count, in what only you can write. And you are only alone on the topics you genuinely practise, which makes your specialization and the quality of your pages inseparable. Length will not save you either: according to the Nielsen Norman Group (2008), during an average visit users have time to read at most 28% of the words on a page, and 20% is more realistic. What they do read has to be worth it.
Google has humans rate that difference. Outside evaluators review search results against a rulebook Google publishes itself, the Quality Rater Guidelines. Section 3.2 covers the quality of the main content, meaning the part of the page the page exists for. Four criteria keep coming back: effort, originality, talent or skill, and accuracy. Section 7.1 describes what is expected of a page rated at the top: main content that took real work and covers the topic all the way. Section 5.2 does the opposite and describes Low quality main content, including content produced without real effort. The full text is online. Here is the consequence that matters: the rating does not stop at the page. A pile of weak pages drags down what Google makes of the whole site.
Effort has to be visible, not announced. Nobody takes your word for it, so the page needs things anyone can see: a number you measured, a photo taken at your own premises, a comparison table you built by hand, a diagram that explains what the text explains badly, a case you handled with its dates and its outcome. A trainer publishing the average scores of 400 past students brings information that exists nowhere else. The same trainer writing "training matters" brings nothing. Readers also judge what they see before what they read: in the Stanford Persuasive Technology Lab study (2002) of 2,684 participants, 46.1% of comments about a site's credibility were about its visual look. On topics touching health, money or law, the ones the manual files under YMYL, the bar rises again: section 3.4.1 puts trust above everything else, and a vague page is not merely average, it is rated low. On the law firm sites I audit, the service page holding one sentence that could not be copied from a competitor is the exception, not the rule.
Answer engines do not cite what they already know. Put a question to ChatGPT, Perplexity or Google AI summaries: none of them has a reason to point at a page repeating what ten other pages already say. They cite the source of a number, a test, a method or a table. An article that summarises current knowledge is exactly what the model produces on its own, in three seconds, without naming you. An agency that publishes the real prices of the 120 quotes it received this year becomes a source: it gets cited because that figure cannot be invented. Now that AI writes competent text on demand, competent text is worth nothing. What is worth something is what somebody had to go out and get.
Google has a name for this: content effort. In the September 11, 2025 PDF of the rater guidelines, the word "effort" appears 120 times, and the page that defines main content quality carries more of them than any other. Cyrus Shepard, who studied this criterion in a 2026 article, gives the most useful definition: effort is not how much work you put in, it is the evidence of useful work the reader can see on the final page. Not word count, not hours spent, not whether a human typed the text. Google's internal documentation that became public in May 2024 points the same way: it holds a field called contentEffort, described as an LLM-based effort estimation for article pages, and a field called ugcDiscussionEffortScore for reader discussions. Nobody knows whether Google uses them or how, and neither is a score you can look up. Keep the principle: a machine can estimate effort by reading the page, exactly as a human rater does. Five questions let you test it yourself. Does the page curate material you actually checked, or copy lists found elsewhere? Is it produced in bulk with one word swapped? Does it hold information that did not exist before you? Does it give the essential in the first lines, or make the reader sit through 600 words on the history of the egg before saying how long to boil it? And if readers discuss on it, does the discussion add anything?
The NeuroEEAT scoring grid (0-10)
Rate what your best page brings that cannot be found elsewhere, not the care you put into writing it.
| Score | Description | Concrete indicators |
|---|---|---|
| 0-2 | Nothing of your own | Text generated or rewritten from the top results. No number, no example, no visual that belongs to you. Remove your logo and nobody would notice the difference. |
| 3-4 | Clean and empty | The text is correct and answers the question, but everything in it already sits on ten other pages. No in-house image, no concrete case, no position taken. |
| 5-6 | A few proofs | One or two contributions of your own per page: an example from your work, a photo you took, a table you built. The rest is still a rehash. |
| 7-8 | Visible work | Every important page carries a number, a test or a case only you have. The topic runs down to the edge cases, price and limits are covered, and the author is identified. |
| 9-10 | A source others cite | Your pages act as a reference: other sites quote them, they publish figures you measure yourself along with the method, and they say when your solution is the wrong one. Updates are dated and explained. |
How to audit it on your site
What to look at on your ten most important pages, not across the whole site.
- 01
Cross out what already exists elsewhere
Take one page. Cross out every sentence whose information sits in the top five results for the same topic. Read what is left. If it fits in two lines, the page brings nothing, however long it is.
- 02
Count the proofs that are yours
On that same page, count the items only you can supply: a measured number, a photo taken at your place, a screenshot of your own tool, a dated client case, a diagram drawn for the occasion. Zero is a common result, and it is the real problem to fix.
- 03
Check that the topic runs to the end
Write down the three questions a client always asks after the main one: what it costs, when it does not work, what happens next. If the page stops before them, it covers half the topic and the reader leaves to finish it somewhere else. Run the reverse test at the top of the page too: how many words must the reader get through before the first useful piece of information? More than one paragraph is filler.
- 04
Look at who signs and what they have done
Is the main content attached to a person, with what they have actually done in the field? A page with no author on a topic that demands skill loses points, even when the text is accurate.
- 05
Spot the pages produced in bulk
Search your own site for pages that match one another word for word except one: same text, city or product swapped. Those are the first an evaluator files under low effort, and they pull the whole site rating down with them.
How to activate or improve it
From your first original contribution to a page other sites quote, one step at a time.
Put at least one proof on every page
- Pick your five most important pages and add one item of your own to each: a photo, a number, a screenshot, a dated example.
- Replace a copied definition with the way you explain it to a client on the phone.
- Answer the question everybody avoids: what it costs, or when you should not do it at all.
- Sign the page with the name of the person who did the work, never "the editorial team".
Take the topic all the way
- Before writing, list the five questions your clients genuinely ask about the topic, and turn them into the outline.
- Add one visual built for this page: a diagram, a comparison table, a before and after. A bought stock image does not count.
- Take a position when it is warranted, and say why. A page that never commits teaches nobody anything.
- Date your figures and say where they come from, with a link to the original source.
- Merge or delete the weak pages covering the same topic: one complete page beats four thin ones.
Become the source others cite
- Publish one dataset a year that only you have: observed average prices, success rates, results of your own tests, with the calculation method.
- Document your test protocols: equipment, duration, conditions. That is what makes a test believable and repeatable by someone else.
- Add the hard part of the topic: exceptions, cases where your advice does not apply, mistakes you made and what they cost.
- Keep an update history: what changed, when, and why.
- Track three numbers each quarter: pages quoted by other sites, pages cited in AI answers, and pages with no original contribution still waiting to be fixed.
5 common mistakes to avoid
Sector specifics
Medical information
A mistake here can harm the reader's health, and section 3.4.1 puts trust above everything else. Effort then also means agreeing with what the authorities in the field say, sources that are current, and an author with the skill. A pleasant text that contradicts medical consensus without justifying it gets the lowest rating in the manual, however carefully it is written.
Comparison site or adviser
Pages comparing loans or investments are almost all built on the same public data. Original contribution is therefore the only playing field left: your own rate readings with their dates, the calculation method written in plain terms, and the cases where your recommendation does not hold.
The city page trap
Spinning one page per city by swapping the name is the most common shortcut and the most expensive. What saves a local page is whatever is only true there: jobs completed in the area, the constraints of local buildings, real lead times, photos taken on site.
Product pages
The manufacturer description sits on every site selling the product. What belongs to you: your photos from several angles, the measurements you took, the questions your customers ask, and what the product does not do. A page that states plainly who the product is wrong for sells better and gets quoted more often.
Signals to activate in parallel
Content effort never stands alone. These signals are what make it visible:
FAQ
How to create quality content with no budget and no team?
Start with what you already own and nobody else has: your files, your job site photos, the questions clients asked you this week. One page a month with a real contribution beats ten pages written on an assembly line. The cost of quality is not money, it is the time it takes to pull out information only you hold.
How many words does quality content need?
No word count is required anywhere in the rater manual. The right length is the one that covers the question and everything that follows from it, then stops. A 700 word page that fully answers a precise question outranks a 2,500 word article circling the topic. Length is a consequence, never a target.
Is AI written content penalised?
The machine is not the problem, the missing contribution is. Evaluators judge the result: effort, originality, skill, depth. Text generated and published as is has none of those. The same text used as a draft, then fed with your figures, your examples and your limits, can be excellent.
What counts as low effort content for Google?
Content where nothing on the page proves a human worked for the reader: a list copied from elsewhere, pages produced in bulk with one word swapped, a summary of what ten other sites already say, filler before the useful information. It is not about length or about the tool you used: a short text written with AI help but fed with your own numbers and photos shows more effort than a long, clean article with nothing new in it. The rater guidelines file that content under low quality, and the whole site suffers when there is a lot of it. The fastest test is still the deletion test: if the page vanished today, what would be missing from the internet?
Do I have to rewrite all my old pages?
No. Sort them into three piles: the ones bringing visits that deserve to be enriched, the ones covering the same topic that are better merged, and the ones bringing nothing that can go. Deleting an empty page often does more good than rewriting it, because weak pages weigh on the rating of the entire site.
Case study
Before · old posts left untouched
After · a historical optimization program
Source: HubSpot (Pamela Vaughan), 2015, https://blog.hubspot.com/marketing/historical-blog-seo-conversion-optimization · Results published by HubSpot on its own blog. The effect comes from posts rewritten in depth, not from a simple date change.