Free AI citability grader

Where the 12 checks and their weights come from

Each check is a plain rule with a fixed weight. A pass earns the full weight, a warning half, a fail nothing. On plain text the checks add up to 94 points and the score is rescaled to 100; paste HTML and the 6-point JSON-LD check joins in. Nothing you paste is sent anywhere.

Three checks carry 12 points each: an answer in the opening, specific numbers, and named, linked sources. Together that's 36 of the 94 plain-text points, close to 40% of the score. The weighting follows the GEO paper by Aggarwal et al. (Princeton and IIT Delhi, 2023), the study that named generative engine optimization. Of the nine rewrites it tested, adding statistics, quotations and cited sources moved visibility in generated answers the most, by up to 40%. Keyword stuffing did close to nothing.

The lighter checks reward passages an engine can lift whole: question headings, short paragraphs, lists, an FAQ, readable sentences.

The exact thresholds behind pass, warning and fail

When a check flags something, this is the line you crossed. These are the rules in the script, nothing hidden.

Which failed checks should you fix first?

Fix the three 12-point checks first: the opening, the numbers and the sources. Count points lost and ignore how many rows are red. A page that fails the three 12-point checks has given up 36 points, and fixing those three usually takes one rewrite of the opening plus an hour finding sources. Five small fails might be worth less than one of them.

1. Delete the warm-up. Your first sentence should answer the H1 in 35 words or fewer: "X is...", "Yes, because...", "It costs...".
2. Add numbers, each linked to where it came from. A number without a source passes this grader and fails a skeptical reader.
3. Rewrite headings as questions people would actually type. "Pricing" becomes "How much does X cost per seat?". Engines split queries into sub-questions (query fan-out), and a heading that matches one gives them a clean passage to take.
4. Leave the rest until those pass. Lists, an FAQ and shorter paragraphs are cheap, but they move the score in 3 and 4 point steps.

Worked example: the same paragraph at 17 and at 78

Here's a typical blog intro on AI Overviews and traffic. Paste it into the grader and it scores 17.

## Introduction

In today's fast-moving digital world, search is changing faster than ever.
Many marketers are wondering what Google's new AI features mean for their
traffic, and the honest answer is that it is complicated and depends on a
lot of different factors, including your industry, your audience and the
kinds of queries that your content tends to rank for over time.

Some studies suggest clicks may be going down, while other people argue
that the traffic that does arrive is higher quality, so it is important to
keep an eye on your analytics and adapt your strategy accordingly.

It fails nine checks. No answer, no numbers, no source, a heading nobody searches for, and sentences averaging well over 26 words. Now the rewrite, using the Pew Research Center study of 900 U.S. adults' Google browsing in March 2025:

## Do AI Overviews reduce clicks to websites?

Yes. In a [Pew Research Center study](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/)
of 900 U.S. adults' Google searches in March 2025, people clicked a regular
result on 8% of visits to a page with an AI Overview, against 15% on pages
without one.

## How big is the drop?

- Clicks on regular results: 15% without a summary, 8% with one.
- Clicks on a link inside the summary: 1% of visits.
- Browsing sessions that ended on the page: 26% with a summary, 16% without.

## Which queries are affected?

About 18% of the 68,879 searches Pew logged showed a summary. Compare
Search Console clicks for those query types over a 3-month window before
you blame AI Overviews for a drop.

That scores 78. Same topic, about the same length. What's still costing points is worth seeing too. The grader reads "Yes." and the Pew sentence as one 38-word opening, over the 35-word limit, so splitting the study details into their own sentence would pass that check. Only one domain is linked, and there's no quote and no FAQ. Getting to 90 would mean adding a second and third source, which is also what would make the claim harder to dispute.

Can a page score 95 and still never get cited?

Yes. The grader checks shape. It can't tell whether your numbers are right, whether the page answers what buyers actually ask, or whether any engine can reach it. A page at 95 that blocks OAI-SearchBot won't appear in ChatGPT search, and that's less rare than it sounds: in our 261-site AI crawler access study, 56 sites blocked at least one AI search bot. Run the robots.txt AI checker before you spend time on wording.

It's also easy to game. Fifteen random percentages and three links to Wikipedia will pass the two heaviest checks. Engines and readers both see through that, and the evidence on how AI engines choose sources points to authority and relevance doing most of the work once a page is in the candidate set.

A score tells you a page is quotable. It doesn't tell you whether engines are quoting it. To see whether ChatGPT and other assistants name your brand for real buyer prompts, the free Arobis AI visibility checker runs that test, and our guide to measuring AI visibility covers how to track citation rate over time. If you need that tracking across several engines every week, our comparison of the best AI visibility tools covers the paid options.

Frequently asked questions