Article · October 21, 2026 · 10 min read
Main content: a self-audit on Google’s four attributes and how to review AI content
In October 2026 Google’s helpful content guide added a definition of main content and named four attributes used to judge its quality: effort, originality, talent or skill, and accuracy. At the same time Google called manual review of all AI content before publishing critical, title, description, markup and alt text included. Here’s how to check your page against these attributes and how to set up the review.
Weak spot: Originality. List what you have that the top 10 doesn’t and move it to the top of the page.
What Google calls main content
Google defines main content as “any part of the web page that directly helps the page achieve its purpose.” That’s not just text. A calculator on a pricing page, a spec table on a product page, reviews, tabs and headings are main content too. We don’t count navigation, sidebars and ads as main content; that’s our reading, Google’s page doesn’t list them.
Page diagram: heading and answer, tool or table, analysis and examples, reviews are main content; sidebars and footer are supplementary; an ad block on the side; navigation on top.
A schematic page. Switch zones: what helps the page do its job, and what supports it.
Google says main content quality is one of the most critical factors in assessing a page. The practical rule: improve what solves the visitor’s task first, not the wrapping around it.
Source: Google: helpful content
The four attributes and how to check each one
The attributes come from Google’s helpful content guide. The “What is judged” column is our paraphrase. The columns on checking and typical failures come from our own practice.
| Attribute | What is judged | How to check your page | Typical failure |
|---|---|---|---|
| Effort | How much human work went into the content or the systems that create it | Does the page have something one AI prompt can’t produce: measurements, calculations, photos, interviews | A rewrite of the first results page |
| Originality | Unique information or perspectives not available on other sites | List three facts or findings that aren’t in the top 10 for the main query | “10 tips” identical to competitors |
| Talent or skill | Whether the content shows the skill needed to satisfy visitors | Are there nuances, mistakes and decisions only a practitioner would know | Generic statements with no detail and no author |
| Accuracy | Factual accuracy on informational pages; a higher bar for money and health topics | Every number, date and price is checked against a primary source, with a review date | Made-up numbers and outdated data |
The attributes don’t substitute for each other: a high score on one doesn’t make up for failing another. A perfectly accurate page with nothing of your own is still a rewrite. The average on the radar at the top is a guide; look at the weakest attribute.
Sources: Google: helpful content, Relevant Audience: authors
Page self-audit
Nine questions. Weight 2 goes to those closest to Google’s attributes: your own data, unique facts, checked numbers. Answer honestly about one specific page, not the site as a whole.
A 15-minute page self-audit
Pre-filled with an example; mark your own page
- The page has data you collected yourselfmustMeasurements, tests, calculations, photos, customer experience.
- It has at least three facts or findings not in the top 10mustCheck the first ten results for the main query.
- It’s clear who the author is and why to trust themName, role, topic experience, link to an author page.
- Every number, date and price is checked against a primary sourcemustNot a summary and not an AI answer.
- The main question is answered on the first screenAn answer, not an intro.
- No paragraph can be deleted without losing meaningFiller and restating the obvious.
- A person reviewed the title, description, markup and alt textmustAnd the markup validates.
- The page isn’t one of a batch of similar AI pagesmustA “one per city” or “one per model” template.
- Changing data shows a review datePrices, timelines, laws, specifications.
Google’s AI content guidelines: how to review
Google explains why manual review is needed: generative models predict a likely sequence of words rather than retrieve facts. So a text can include plausible but false statements. We run the review in this order:
- Separate claims from the textfacts
List every number, date, price, name, product property and promise.
- Check against a primary sourcefacts
Manufacturer documentation, the law, official statistics, your own data. A summary or an AI answer is not a source.
- Remove what didn’t check outfacts
Don’t soften it; delete it or replace it with something verified.
- Add your own contributionvalue
Data, experience and examples competitors don’t have. Without this the text is still a rewrite.
- Check the metadatametadata
Title, description, alt, structured data. Validate the markup.
- Say how the content was madetransparency
Where readers would care: “prepared with AI, reviewed by a named specialist”.
- Final read by a personsign-off
The person responsible for the topic reads the whole page as a visitor would.
Source: Google: generative AI
Metadata: what to check
Metadata appears in search results, so Google applies the same manual review rule to it. The typical errors below come from our practice.
| Element | What usually breaks | How to check |
|---|---|---|
| Title | Promises the page doesn’t keep; identical titles across dozens of pages | Compare with the H1 and first screen; find duplicates with a crawler |
| Meta description | Made-up numbers and benefits, template filler | Every number in the description appears on the page |
| Structured data | Ratings and reviews that aren’t on the page; wrong prices and dates | A markup validator plus checking values against visible content |
| Alt text | Describes what isn’t in the image; keyword lists | Open the image and read the alt next to it |
| Product data | Generated specifications without a label | Google Merchant Center rules require AI-generated product data to be specified separately and labeled |
For ecommerce, Google Merchant Center rules require AI-generated images to carry IPTC DigitalSourceType metadata set to TrainedAlgorithmicMedia, and generated product titles and descriptions to be specified separately and labeled as AI-generated.
Source: Google: generative AI
Where AI help ends
Google isn’t against AI as a tool. The line is drawn by purpose and result. The scaled content policy describes a violation as many pages made mainly to manipulate search rankings with little value for users.
- Many pages generated with AI without added value.
- Stitching together others’ content without your own contribution.
- Pages that make no sense to a reader but contain keywords.
- Several sites used to hide the scale of generation.
A simple test for your own site: if you removed every page in a section that was produced in a batch in the last six months, would visitors be worse off? If not, those pages are candidates for merging or noindex. How to rebuild such a section into a structure where every page has its own job is in part 4.
Sources: Google: spam policies, Google: helpful content
FAQ
What is main content according to Google?
Any part of the page that directly helps it achieve its purpose: text and media, interactive features, user contributions, tabbed sections, titles and headings.
Which attributes does Google use to judge content quality?
The helpful content guide names four: effort, originality, talent or skill, and accuracy. Search quality raters use the same attributes.
Do I need to review AI content manually?
Yes. Since October 1, 2026, Google calls manual fact-checking and review of all AI content before publishing critical. This covers metadata too: titles, meta descriptions, structured data and alt text.
Should I disclose that AI was used?
Google recommends explaining how content was created where readers would reasonably wonder. For ecommerce, Google Merchant Center rules require AI-generated product data to be specified separately and labeled.
What’s the risk of mass-generating pages?
If many pages are made mainly to manipulate rankings with little value for users, that violates Google’s scaled content abuse policy.
Sources
Definitions and requirements are from Google Search Central documents checked on October 17, 2026; Google’s wording is paraphrased except where quoted. The self-audit questions, review order and typical errors are our method. The radar and checklist are pre-filled with an example.
How this piece was made: written by the SurfUp editorial team with AI assistance; every fact and number was checked by hand against the sources listed above.
- Google Search Central — Creating helpful, reliable, people-first content (updated October 5, 2026)
- Relevant Audience — Google adds fake-author warning to helpful content page
- Google Search Central — Guidance on using generative AI content (updated October 1, 2026)
- Google Search Central — Spam policies: scaled content abuse