Score your article before you publish it

What the content score measures

It judges whether your page deserves to rank for the keyword you give it, compared with the kind of pages that already rank for that keyword. The keyword is required because the judgement is about this page for this query. Score the same text against another keyword and the result can change.

The result is one of four bands. Weak means the page is below what ranks now and needs a rewrite before it can compete. Fair means it is publishable but would sit below the current top results. Good means it is competitive with them. Strong means competitive and leaning toward beating them.

The band comes from one overall judgement of the page. It is not an average of the lines under it, so a Good page can still show a weak line.

How a score is produced

First the paste is checked. It needs a keyword, at least 300 words and no more than 12,000 characters, markup and spaces included. A paste that fails a check is turned away before it uses any of your free scores.

Words are counted on the prose alone. Code blocks, Markdown table rows, images and link addresses are dropped before counting. Heading text still counts.

Then one AI model reads the article as you pasted it, Markdown and HTML included, because the structure line judges headings, lists, tables and FAQs. It answers ten questions: one overall judgement for the keyword, six quality axes, where the answer sits, and two checks for specific habits.

Each axis answer is rounded to the nearest of five levels, and you see the line written for that level, weakest first. A habit is reported only when the model rates it at least as likely as not.

No numbers are shown. The free result is the band, seven lines and any habit that fired.

What each line in the result means

Six lines are quality axes and the seventh is answer position. Each comes back at one of five levels, shown by the bar beside it.

• Writing: the prose itself. Varied sentences and paragraphs rate high, stock transitions and paragraphs cut to one shape rate low. Whether the content is right or complete is left to the other lines.
• Coverage: whether the page covers what a knowledgeable reader expects for the keyword, without leaving out a main meaning of the query.
• Information gain: how far the page commits to specifics, such as real figures, comparisons a reader can act on, where the usual advice stops working, or a position it defends. Restating what every other page says rates low.
• Evidence: whether the claims are right and specific. The model is told not to mark a page down for missing links or citations. The bottom level describes claims that contradict each other, or testing and ownership the page implies but never establishes.
• Structure: whether every heading, list, table and FAQ is there because the material needs it. Scaffolding heavier than the content rates low, and a tight short page can rate above a long segmented one.
• Spam safety: whether the page reads as editorial or as templated output, such as points repeated in new wording, a recap closing each section, or an FAQ that exists to rank. A higher level is safer.
• Answer position: where the direct answer to the keyword first appears. At the top level it is the first sentence. At the bottom it is never stated outright.

The two habit checks sit apart from the lines. One looks for an FAQ that exists because FAQs rank, with questions that restate the body or show what the article never researched. The other looks for hands-on experience the article implies, such as testing, owning or visiting, with no account of when or how. When neither fires, the result says so.

What to do with a low score

Start at the top of the list. The lines are sorted weakest first, so the first one or two show where the page falls shortest.

If answer position is low, put the direct answer to the keyword in the first sentence, before any background. If information gain is low, add what most pages on the topic leave out: a real figure, a comparison, a case where the usual advice fails, or a position you can defend.

If evidence is low, swap vague figures for exact ones and fix numbers that contradict each other. If the page implies testing, owning or visiting, say when and how it happened, or stop implying it.

If structure or spam safety is low, cut before you add. Take out the recap at the end of each section, the takeaway box and any FAQ that repeats the body.

If writing is low, vary sentence and paragraph length and drop the stock transitions. If coverage is low, cover the main thing people searching that keyword want before the side topics.

Then score it again. Changed wording counts as a new text, but every run still counts toward your daily cap.

Raising one line does not guarantee a higher band, because the band is a separate judgement of the whole page.

What the free score does not do

It does not look at search results. No results page is fetched and no competitor is crawled, so when it compares your page with what ranks, it is going on what the model expects to rank for that keyword.

It does not check facts against any source. The evidence line judges whether claims are specific and consistent with each other, and the model is told to treat a specific, dated event it does not recognise as true.

It does not say whether a person or a model wrote the text, and it does not look for copied text.

It reads only what you paste. It does not open links, and a paste over 12,000 characters is refused, not cut.

It gives levels, not numbers, and quotes no passages. One model reads the page once, and search intent, which the full report rates, is not part of the free score.

Free use has a daily cap per visitor, per network and per text. The counters reset at midnight UTC, and the scorer as a whole has a daily cap too.

The free score and the full report

The paid Article Quality Report goes further. Two or three models you pick rate the article separately, each on seven axes from 0 to 10, search intent included.

Each model quotes the passages that earn or cost points, says what to change, and writes a short verdict. The report shows the average for each axis and how closely the ratings agree.

It runs from your dashboard and is paid in credits. New accounts start with 5,000 free credits, no card needed.