How we score an article
Last updated:
Short answer: the article checker applies fixed, published rules, and every one of them is listed here with its basis. There is no hidden model and no benchmark against other people's pages. Overall is 40% SEO, 30% AEO and 30% GEO. A score is a readiness estimate, not a prediction of rankings or citations.
What the three scores mean
- SEO covers the writing itself: headline, headings, where your phrase appears, length, ease of reading and images. On a published page it also covers the Google preview text and the Article label. Speed, security and whether Google can reach the page belong to the website checkup.
- AEO asks whether Google's answer box, or an AI assistant, could take a direct answer straight out of it. It looks for an answer up front, headings written as questions, lists, tables, plain definitions and a summary.
- GEO asks whether an AI assistant such as ChatGPT, Gemini or Perplexity would cite it. It looks for statistics, sources, quotations, a named author and date, first-hand experience, and specifics instead of adjectives.
The sources
Two published documents anchor most of the rules.
The first is Google's own guide to AI features. It says a unique point of view and first-hand experience beat writing anyone could have produced. It says to write for people, with clear headings and sections. In its own words, you “don't need to create new machine readable files, AI text files, markup, or Markdown” to appear in Google Search.
The second is the KDD 2024 study by Aggarwal and colleagues. They measured what actually moves the number, and got visibility up by as much as 40%. Their three best changes were adding statistics, adding quotations, and citing sources. Readability helped, but less. Keyword stuffing did nothing, and on one engine it did 10% worse than making no change at all. That is why no rule here rewards repeating your phrase. We only check that it is there.
- Google, Guide to AI features in Search
- Google, Creating helpful, reliable, people-first content
- Google, Influencing your title links in search results
- Google, Control your snippets in search results
- Google, Article structured data
- Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024
- Flesch reading-ease formula (Flesch, 1948), as used by most readability tools
SEO rules (16)
| Rule | Weight | What passes | Basis |
|---|---|---|---|
| A headline | high | The piece has one clear headline. | |
| Headline length | medium | 30 to 60 characters (20 to 70 is a warning). | |
| One main headline | medium | Exactly one main headline (Heading 1); sections use Heading 2 and Heading 3. | |
| Section headings | medium | Two or more section headings, and no Heading 3 straight after the main headline (from 600 words). | |
| Your phrase in the headline | high | The phrase you want to be found for appears in the headline. | |
| Your phrase in the opening | medium | The phrase appears in the first 100 words. | |
| Your phrase in a section heading | low | The phrase appears in at least one section heading. | |
| How often your phrase appears | medium | 0.3% to 2.5% of all words; over 3.5% fails as overuse. Counted as whole words. | Aggarwal et al. |
| Length | high | 600 words or more (300 to 600 is a warning). | |
| Easy to read | medium | Reading ease of 50 or more out of 100 (30 to 50 is a warning). | Aggarwal et al. |
| Paragraph length | low | Average under 120 words and none over 200. | |
| Google preview text | medium | Published page only: the short text under your link in Google is set, 70 to 160 characters. | |
| Links to your own pages | low | Published page only: two or more links to other pages on your site. | |
| Article label for Google | medium | Published page only: the hidden Article label (structured data) is present. | |
| Image descriptions | low | Every image has a description (alt text). | |
| Images | low | At least one image (not judged for pasted text). |
AEO rules (10)
| Rule | Weight | What passes | Basis |
|---|---|---|---|
| Answer in the first paragraph | high | The opening paragraph is 20 to 90 words and answers the question. | |
| Enough detail to answer | high | At least 300 words; below that the piece cannot be the best answer. | |
| Headings written as questions | medium | At least one section heading ends in a question mark, or opens with how, what or why. | |
| Lists or steps | medium | At least one bulleted or numbered list (from 300 words). | |
| A table for comparisons | low | At least one table (from 300 words). | |
| A one-sentence definition | medium | At least one sentence that defines the subject, such as “A felt roof is …”, anywhere in the piece. | |
| Questions and answers | low | A questions heading, or three or more question headings. | |
| Sentence length | medium | Average of 20 words or fewer (up to 25 is a warning). | Aggarwal et al. |
| Summary or key points | low | A summary, key points, TL;DR or short-answer section. | |
| Paragraphs short enough to quote | low | Half or more of paragraphs are 40 to 120 words (from 300 words). |
GEO rules (10)
| Rule | Weight | What passes | Basis |
|---|---|---|---|
| Numbers and facts | high | Three or more numbers with units, percentages, prices or counts per 1,000 words. A bare year does not count. | Aggarwal et al. |
| Links to your sources | high | Two or more links to other websites, or phrases such as “according to”. | Aggarwal et al. |
| Quotations | medium | At least one direct quotation or block quote. Quoted code and settings do not count. | Aggarwal et al. |
| Author named | medium | An author line, document author or author label is present. | |
| Published or updated date | medium | A date is present. | |
| Your own experience | medium | At least one first-hand phrase (we tested, our data, case study, and so on). | |
| Why the author is qualified | low | A qualification is stated (doctorate, chartered, years of experience), or the named author's role is given. | |
| Facts instead of buzzwords | medium | Three or fewer vague words or buzzwords per 1,000 words (over six fails). Counts unbacked superlatives (“leading”, “world-class”) and long words where a short one would do (“leverage”, “utilise”). | |
| Same subject start to end | low | The main word of your phrase, or of the headline, appears in both the opening and the closing third. | |
| A recent year mentioned | low | This year, last year or next year is mentioned. Judged against today's date. |
How the numbers are made
Each rule has a weight: high counts 5, medium 3, low 1. A pass earns the full weight, a warning half, a fail nothing. Rules that cannot apply to what you gave us are left out entirely, such as a Google preview text when you paste a draft. A category score is the share of the weight you earned. Thin or vague writing can score well on one of the three and badly on another. That is deliberate.
What we do not do
- We do not store the text you check. We keep the scores, the pass or fail outcomes, the word count, the keyword and, for a URL, its address. That is how we see which problems are most common.
- We do not send your draft to an AI model. Every rule runs on our own server.
- We do not compare you with other people's pages, and we do not claim to know what will rank.