AI writing is embarrassing to send. What you don't see is worse.

Blue Pencil checks every draft against your own definition of a good document and scores each rule. You see what to fix before you hit send.

Try it live. Nothing to install.

You've sent one of these.

AI drafts are embarrassing, and some are wrong. You asked AI for a quick update to a client, and it wrote three paragraphs that start with "I hope this finds you well" before the one line your client needed. You caught that one.

The ones you don't catch are worse: a date your team never agreed to, a bug called fixed when your team has only found it, "no data was lost" turned into "your data is completely safe". Your client reads each of those as a promise.

Reading every AI draft line by line takes time. Two in three US workers say they don't always check what AI gives them (Kolmogorov Law, a September 2026 survey of 500). So drafts go out on hope.

Measure a draft against your own standard.

Write down what a good document of your kind needs: its parts, and the rules each part follows. Blue Pencil calls that list a schema.

Blue Pencil then checks your draft against every rule and scores each one from 0 to 1. A rule passes at 0.5. You see each failed rule, and where it failed.

Slop or Not

Here is one AI email, before and after Blue Pencil. It comes from a real Blue Pencil run: Blue Pencil failed the first draft on three checks and was unsure on two more, and it passes the fixed version on all 62. At Spectacular AI, the team that makes Blue Pencil, we gave an AI four facts and asked it for a short customer email about a delayed release. Below is its first draft, unedited.

Before

Subject: v2 Release Update – October 15 Launch Hi there, We're postponing our v2 launch from October 1 to October 15 to ensure quality. Our testing team discovered a bug in the data export feature that we want to fix before release. If you're currently on v1, there's nothing to worry—you're unaffected and can continue working normally. Your data remains completely safe. We appreciate your patience as we make v2 even better. Thanks for being part of our journey. Best regards, [Your Team]

What Blue Pencil found

Scores run from 0 to 1, and a check passes at 0.5. Between 0.2 and 0.8 a check is unsure, so you read that part yourself.

No stock phrases from other company emails:0.00,failed.
No em dashes:0.16,failed.
The main point comes first:0.47,failed.
No claim beyond the four facts:0.59,unsure.
Says that no customer data was lost:0.66,unsure.

The two unsure checks are right. The facts say no data was lost, not that it's "completely safe". v2 was delayed to fix a bug, not to make it "even better".

After

Subject: v2 now ships on October 15 Hi, v2 will ship on October 15 instead of October 1. Testing found a bug in data export, and we want it fixed before release. If you use v1, nothing changes for you. No customer data was lost. [Your Team]

Three steps, and Blue Pencil does the first one for you.

Describe what good looks like: paste a document you already have, and Blue Pencil writes your first rules from its headings.
Check a draft: paste it in to get a score for every rule.
Fix what scores low, then check again.

We run it on our own docs.

At Spectacular AI, we use Blue Pencil on the notes that document our own code. We cut wrong or unsure checks from 31 of every 100 to 4 in three rounds of rewriting the rules. You can read how we did it in our case study.

314

Checks wrong or unsureof every 100

From 31 wrong or unsure checks in every 100 to 4: how we tuned Blue Pencil for our own technical notes

This is how we at Spectacular AI use Blue Pencil internally, on the technical notes that document our own code.

Fair questions

Isn't this just AI grading AI?: Yes, which is why each check is narrow. Each check is one question about one rule, such as "does the email say when v2 ships?", with a score for its answer. When a check gets split answers, Blue Pencil gives it a score between 0.2 and 0.8 and marks it unsure, so you know where to look. On Spectacular AI's own notes, 96 of every 100 checks were right and clear.
How much work is it to start?: It takes one paste. Blue Pencil writes starting rules from your document's headings and checks the document against them. You change the rules whenever you like.
What happens to my text?: Blue Pencil keeps nothing. It sends your text to an AI model for scoring and stores none of it after the review.
What does it cost?: Trying it live is free while Blue Pencil is in preview.
I skim my drafts, and they're usually fine.: "Usually" is the problem. At skimming speed you can't tell a bad draft from a good one, and the mistake that matters is the one you skimmed past. Blue Pencil checks every rule on every draft, even when you don't have time to.

Try Blue Pencil live on a draft you wrote with AI this week.

Paste it in, and in seconds you'll see a score for every rule on your own document.

Try it live. Nothing to install.