Selected · 9 of 100 points

# Does formatting change whether AI quotes your page?

It does: lists and tables keep their row and column relationships when a page is chunked, and the same content written as flowing prose loses them the moment it is extracted.

[Score my page on this](/tools/geo-score-checker)[See all 14 checks and their weights →](/checks)

- 9 points of 100
- 9/60 of the Selected lever
- scannability check id

By [Iliass El Barhoumi](/about) · Published 1 October 2026 · Last updated 1 October 2026

## How is scannable formatting scored?

Scannable formatting is worth **9 of 100 points**, 9 of the 60 in the Selected lever, which asks “Is there a passage worth quoting?” The engine scores it with these rules:

- 3 points for at least one bulleted or numbered list with two or more items. Paragraphs that enumerate in prose (“first… second… finally…”) are named in the fix text.
- 2 points for at least one table.
- 2 points when 70% of paragraphs are 80 words or fewer.
- 2 points for at least 2 emphasised key terms: bold, mark or inline code.

## What does a pass and a fail look like?

| Verdict | Example                                                                                                                                                                                              |
| ------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Pass    | A comparison of three espresso grinders as a table with columns for price, burr type and grind settings. Each row survives extraction with its column labels.                                        |
| Fail    | The same comparison as a paragraph: “The first grinder costs more but has conical burrs, while the second, which is cheaper, uses flat burrs…”. After extraction, nothing ties a price to a grinder. |

## How do you fix scannable formatting?

In the order worth doing it:

1. Turn any sentence that counts (“first”, “second”, “finally”) into a numbered list.
2. Turn any comparison of two or more items across the same attributes into a table.
3. Bold the two or three terms a reader should remember from each section, and no more.

A scan reports the evidence behind the verdict, such as the sections, paragraphs or tokens that failed, and the fix prompt hands the same evidence to a coding agent so it can make the change for you.

## Related checks in the Selected lever

- [What is answer-first writing?](/checks/answer-first) — 13 points
- [What makes a passage extractable by an AI engine?](/checks/extractability) — 10 points
- [Do headings matter for AI search?](/checks/headings) — 9 points
- [Why do AI engines prefer pages with numbers in them?](/checks/specifics) — 8 points
- [Does an FAQ section help you get cited?](/checks/faq) — 7 points
- [Which schema types help with AI citation?](/checks/schema) — 4 points

## Frequently asked questions

- ### Do AI engines read HTML tables?  
Yes. Tables are parsed as structured rows, so a value stays attached to its row and column labels when the table is chunked, which prose cannot guarantee.
- ### Can a page have too many lists?  
For readers, yes. The check rewards having structure where the content is structured; it does not award more points for more lists, so converting real prose into bullets gains nothing.

## Fix the page, then watch AI arrive

A better score makes a page eligible to be quoted. CrawlRadar then shows whether it happens: the AI crawlers that fetch the page, and the visitors ChatGPT, Perplexity and Claude send to it.

Or start with the free [AI crawler checker](/tools/ai-crawler-checker).

---

Source: https://www.crawlradar.xyz/checks/scannability
