Most advice on how to improve readability stops at “write shorter sentences” and leaves it there. This is the scoring sheet behind that instruction, with the point values attached.
We score every client blog post out of 100 before it earns its keep. Across the last twelve articles we scored on one B2B estate, one check cost more points than any other, on every single article: reading ease.
Those twelve lost 108 points to reading ease alone. Sentence length took another 54. No article escaped either.
Below is the full rubric, the four checks that cost the most, and what to do about each one.
What is a readability score?
A readability score estimates how much effort a piece of writing takes to understand. The most widely used one is the Flesch Reading Ease formula, published by Rudolf Flesch in 1948 and still built into Yoast, Semrush and Microsoft Word today.
It runs from 0 to 100, and it counts two things: how long your sentences are, and how many syllables your words have. Nothing else.
| Score | Band | What it reads like |
|---|---|---|
| 90-100 | Very easy | Understood by an average 11-year-old |
| 60-70 | Plain English | The target for most business writing |
| 30-50 | Difficult | Academic and legal register |
| 0-30 | Very difficult | Best understood by graduates |
A score of 60 to 70 is the working target. Below 30 means a reader has to work, and most will not.
Why readability decides whether AI engines quote you
Here is the part that changed recently.
When ChatGPT, Perplexity or Google’s AI Overviews answer a question, they lift a passage from a source and reuse it. A long, qualifier-stacked sentence is a bad passage to lift. A short, direct one is a good one.
The most-cited research on this is the GEO paper (arXiv 2311.09735), which tested which content changes actually increase how often generative engines cite a source. Three came out on top:
Three of those lift citations. The fourth is the only change tested that made things worse.
Every one of those winners is a writing change. None of them is a technical change.
Microsoft’s own guidance for AI search points the same way. It recommends “clear headings, tables, and FAQ sections” - the content, not the markup behind it.
How to improve readability: the four checks that cost the most
These are ranked by how many points they actually cost across the twelve articles we scored, not by how often they get talked about.
1. Reading ease - 12 points, and the biggest single loss
All twelve articles scored poorly here. The pattern was identical every time: long clauses, stacked qualifiers, and lists run into paragraphs instead of being set as lists.
The fix is mechanical, which is why it is the best value on the board.
Before: “Survey feedback, stakeholder engagement, qualitative comments, escalation themes, exception patterns, response quality and post-move insight can all help show where confidence is strengthening and where it may be weakening.”
After: “Several signals show where confidence is strengthening or slipping:” followed by those seven items as a bulleted list.
Same information. One dense 34-word sentence becomes a six-word lead and a list a reader can scan.
2. Sentence length - 8 points
The target is under 25 words. Not every sentence, but most of them.
Nielsen Norman Group’s research on how people read online found that roughly four in five users scan rather than read word by word. Their study measured each writing change separately: concise writing tested 58% more usable, scannable writing 47%, and objective rather than promotional language 27%. A version combining all three measured 124% more usable, which is the figure usually quoted without the condition attached to it.
Two habits recover most of this:
- Split any sentence containing “and”, “which” or “including” in the middle
- Set any run of three or more items as a list
3. Question-format headings - 7 points
Eleven of our twelve articles had none.
An AI engine matching a user’s question to a source looks for a heading that resembles the question. “Measurement approach” does not resemble anything anyone types. “What should mobility data actually measure?” does.
Write the heading as the question your reader would ask out loud, then answer it in the first sentence beneath.
4. A real FAQ section - 6 points
Also missing from eleven of the twelve.
Three to five genuine questions with direct answers, placed at the end. Pick questions people actually search for rather than questions you wish they asked. A keyword tool will tell you which is which in about a minute.
What about FAQ schema?
This one needs care, because the ground moved and a lot of published advice has not caught up.
Google removed the FAQ rich result for every site on 7 May 2026 and took the documentation down on 15 June. If you added FAQPage markup to win those dropdowns in the search results, they are gone.
That is not a reason to strip it out. Google’s own AI optimization guidance says structured data is not required for AI features, and then continues: “However, it’s a good idea to continue using it as part of your overall SEO strategy, as it helps with being eligible for rich results on Google Search.” The second half of that sentence gets quoted far less often than the first.
Three things are true at once, and most advice picks one and drops the others.
- No search vendor has ever stated that its AI reads FAQ markup. OpenAI’s, Anthropic’s and Perplexity’s crawler documentation mention schema.org zero times.
- The GEO paper never tested schema at all. Search the full text for “schema”, “markup” or “structured data” and you get nothing. It tested writing changes.
- The FAQ content is what does the work. The markup is worth keeping for ordinary SEO reasons, and worth nothing as an AI-citation play.
Build the FAQ section, because the content is what gets quoted. Keep the markup, because it still counts for ordinary SEO. Do not expect the markup itself to earn a citation.
The full scoring sheet
Four pillars, 100 points. This is the sheet, with what each check is worth.
Readability - 30 points
| Check | Points | What good looks like |
|---|---|---|
| Reading ease | 12 | Plain enough to read without effort |
| Sentence length | 8 | Most sentences under 25 words |
| Paragraph size | 6 | No walls of text |
| Passive voice | 4 | Mostly direct, active sentences |
Keyword targeting - 25 points
| Check | Points | What good looks like |
|---|---|---|
| Keyword in title and H1 | 7 | The main term appears in both |
| Keyword used early | 4 | It appears in the first 100 words |
| Keyword density | 6 | Natural throughout, never stuffed |
| Keyword in headings | 4 | At least one heading carries a target term |
| Title and meta length | 4 | Neither gets cut off in search results |
AI-search readiness - 30 points
| Check | Points | What good looks like |
|---|---|---|
| Heading structure | 5 | One H1, a logical hierarchy beneath |
| Question-format headings | 7 | Headings phrased as questions people ask |
| FAQ section | 6 | A real FAQ block at the end |
| FAQs backed by demand | 6 | The questions match real search volume |
| Answer-first | 6 | Each section opens with the direct answer |
Authority and trust - 15 points
| Check | Points | What good looks like |
|---|---|---|
| Named author | 5 | A real byline, not anonymous |
| Publish and update dates | 5 | Both visible on the page |
| Outbound citations | 5 | Links to reputable external sources |
Bands: 80 and above is publish-ready. 60 to 79 needs specific fixes. Below 60 needs rework before it goes out.
Where the checks come from
None of this is a house style invented from scratch. Each pillar maps to a published standard.
- Readability uses the Flesch Reading Ease formula, the same measure inside Yoast and Semrush
- Structure follows Nielsen Norman Group research on scanning behaviour
- AI-search readiness follows Google’s structured data guidance and the tested findings in the GEO paper
- Authority follows Google’s E-E-A-T quality framework
Frequently asked questions
What is a good readability score?
Between 60 and 70 on the Flesch Reading Ease scale for most business writing. That reads as plain English. Technical audiences tolerate 50, but below 30 you are losing readers who could have understood you.
How do I improve readability on a page that already scores badly?
Shorten sentences and break lists out of paragraphs. Those two changes recover more points than anything else, because sentence length is one of only two inputs the Flesch formula uses. Cutting a 34-word sentence to two 17-word sentences moves the score immediately.
Does readability affect SEO?
Indirectly, and increasingly directly. Google has never confirmed a readability ranking factor. What is measurable is that scannable writing keeps people on the page, and that AI engines quote short, self-contained passages far more readily than dense ones. As answer engines take more of the result page, how quotable a sentence is starts behaving like a ranking input.
Which readability formula should I use?
Flesch Reading Ease is the most widely supported, so it is the one most tools report and the one worth optimising against. Flesch-Kincaid Grade Level uses the same two inputs and expresses the result as a US school grade instead of a 0 to 100 score.
Can I just run my content through an AI to fix it?
For sentence splitting and list formatting, yes, and it does that well. What it will not do is know which questions your buyers actually search for, which is where the FAQ points and the keyword points live. That part needs demand data.
Scoring a page against this sheet
Every score in this piece came from the same scorer we run against client blogs each month. The four pillars above are the whole rubric, so the scoring is reproducible: paste the checks against your own post and you will get the same number we would.
The scorer is published as a free tool: paste a URL into the content scoreability checker and it returns the score with every criterion shown, so a deduction always names the pillar that lost it. No signup. The write-up of what it does and does not measure is here.
If a post scores below 60, start with the sentences. It is the cheapest fix, it moves the largest single number, and on every article we have scored it was the biggest loss on the board.