Why this signal matters for Google
Fact checking means walking a claim back to whoever produced it. Take a number everyone in your industry repeats. You click the link in the article quoting it. It points to another article. Which points to a blog post from 2016. Which points to nothing. At that moment you do not have a fact: you have a well-dressed rumour. The rule fits in one sentence. Until you have the original document, its author and its date, the number stays out of your page. And once you have it, you link to that document, the way the guide on cited sources describes. Nobody sees the checking you did. Everybody sees the source.
Human evaluators look at whether your site has a reputation for getting things right. Google has people rate pages using a public manual, the Quality Rater Guidelines (QRG for short). Section 3.2 tells them to research what others say about you, away from your own site: press, forums, reviews, conversations between professionals. A site known for errors pays for it right there. Picture a local news site that announced three shop closures that never happened. Its next articles are spotless, but its reputation is already written elsewhere, and those are the pages an evaluator reads.
On sensitive topics, a mistake does not cost the same. When a page deals with health, money or safety, Google files the topic under YMYL, for Your Money or Your Life. Section 3.4.1 then expects real trustworthiness, and content that contradicts what the specialists in the field say can be rated at the very bottom of the scale. In practice: writing that a supplement replaces a treatment, or quoting a tax rate that changed last year, is not rescued by good design. The reader can act on it and lose money or health, and that risk is exactly what Google is measuring.
AI tools copy your numbers, and check them by cross-referencing. When ChatGPT, Perplexity or Google AI summaries answer a question, they look for several sources saying the same thing. A lone number, with no date and no source, cross-references with nothing, so it gets dropped. A dated, sourced number that others have picked up becomes quotable. The reverse happens too. An error published on a well-ranked page gets copied into AI answers for months, with your name next to it. A dated content update on the day a number changes is the only way to stop that spread.
The NeuroEEAT scoring grid (0-10)
Rate what a reader can check for themselves on your page, without taking your word for it up front.
| Score | Description | Concrete indicators |
|---|---|---|
| 0-2 | Nothing checkable | Claims and numbers with no source at all. No dates anywhere. Nobody can tell where any of it came from, or whether it is still true today. |
| 3-4 | Decorative sources | A few links, but to other articles that are themselves quoting someone else. Numbers carry no date. Old pages are never reread. |
| 5-6 | Right on the essentials | The important numbers have an original source and a year. Sensitive pages get reread once a year. Nothing is written down: it depends on who is writing. |
| 7-8 | Checking is organised | Every number points to the original document, with its date. A review is scheduled. Corrections are made and flagged on the page concerned. |
| 9-10 | Checkable end to end | A named person signs off the facts before publishing. Corrections are dated and explained in public. Sensitive pages are reviewed by someone from the field, and your own data is published with its method. |
How to audit it on your site
Five checks to run on your pages, starting with the ten that get the most visits.
- 01
Spot the sentences that can be wrong
Highlight every number, date, price, rule and recommendation. Those are the only sentences that can be fact checked. The rest is your opinion or your experience, and that is handled differently: by presenting it plainly as yours.
- 02
Walk every number back to its original source
Click the link you are quoting. If it leads to an article quoting another article, keep going. Stop only at the original document: the study, the official report, the manufacturer page. If the chain breaks along the way, the number comes out of the page.
- 03
Write down the year of every fact
A 2019 number is not wrong, it is old. Note the publication year next to each source. Anything older than three years on a moving subject, such as prices, rules or software, gets rechecked before it stays online.
- 04
Compare with what the specialists say
For every recommendation, check that a recognised source in the field says the same thing: a public agency, a professional body, the official documentation of the tool. If you are the only one in the world writing it, either you have data of your own and you show how you got it, or you are wrong.
- 05
Look at what happens after a mistake
Search your own site for a visible correction. Is there a dated line saying what changed and why? If you fix things silently, the reader who saw the wrong version has no way of finding out, and you lose the credit the correction should have earned you.
How to activate or improve it
Three levels, from checking by hand to a method you can show anyone.
Check before you publish
- Before publishing, reread the page and highlight every number, date and price.
- For each one, open the original source and compare the figure word for word.
- Put the year in the sentence itself: "according to a 2024 study".
- Delete anything you cannot trace, however good the line sounds.
- Link to the original document, not to the article that summarised it.
Make the checking visible
- Show two dates in plain sight: published, and last checked.
- Add a short note at the foot of the page saying where your numbers come from.
- When you fix an error, write the correction and its date on the page.
- Schedule a quarterly reread for the pages where numbers move.
- Keep the source list for each page, so you know what to recheck in five minutes.
A method you can show
- Publish how you check facts: who checks, at what point, and against what.
- Have sensitive pages reviewed by someone from the field, and name them on the page.
- Keep a page of dated corrections, linked from the articles concerned.
- For your own data, publish the method: who was surveyed, when, and how you calculated.
- Track two numbers each quarter: errors reported, and the delay between report and fix.
5 common mistakes to avoid
Sector specifics
Consensus comes before personal opinion
A topic with real risk for the reader, where section 3.4.1 expects real trustworthiness. Every sentence about a treatment, a dose or a symptom comes from a recognised medical source, with its date. A lived case can illustrate, never prove. And when the specialists disagree among themselves, say so, instead of picking the side that suits your argument.
The numbers that change every year
Rates, caps, thresholds, deadlines: these move often. The real risk is not a maths error, it is the 2023 page still online with 2023 figures. Put the year in the sentence itself and book the reread for the moment the new numbers come out.
Practical details first
Here accuracy is about opening hours, the address, the prices shown and the area you serve. A customer who drives over for nothing remembers that mistake better than any sales argument. On the same day, check that your site, your business profile and your directory listings all say the same thing.
Official documentation wins
A tutorial written for version 3 of a tool becomes wrong the day version 4 ships. Note the version you tested and the date of the test in the article. When you describe what the product does, link its official documentation rather than a third party post that may have misread it.
Signals to activate in parallel
Accuracy never proves itself alone. These signals show the reader that you actually check:
FAQ
How to fact check when you are not a journalist?
By walking the chain back, which takes no training at all. Click the source quoted, then its own source, until you reach the original document: study, official report, manufacturer documentation. Note its author and its year. If the chain stops at an article that quotes nobody, the claim has no starting point, and you do not publish it.
Does every sentence need a source?
No. Only checkable claims do: numbers, dates, prices, rules, recommendations. Your experience and your opinions have no source to quote, they need to be presented plainly as yours. Too many links make a page painful to read, just as no links make it impossible to check.
What do you do when you find an error on an old page?
Fix it, then say so. Change the text, add a dated line stating what changed, and update the last checked date. Fixing it silently solves the problem for the next readers and leaves the previous ones with the wrong version. A visible correction earns credibility, it does not cost any.
How do you find out what the specialists in a field say?
Work out who holds the authority and go straight to them: public agencies, professional bodies, official documentation, trade journals. Three independent sources saying the same thing beat ten articles copying each other. If nobody else is writing what you are writing, that is not an exclusive, it is a warning.
Can content written with AI be accurate?
Yes, as long as you check every fact yourself. These models produce numbers and references that look right and do not exist. Treat any AI output as an unchecked draft: every figure and every citation goes through the original source before publishing. The responsibility stays with the name signing the page.
Anonymized case study
Before · nothing to trace
After · every number leads to its document
Illustrative figures meant to show the typical progression of the signal. They do not come from a real client case and must be replaced with your own measurements before any external communication.