About Content Trace
Built by Colin H at Web Thrive · Read our manifesto
Why we built this
AI writing tools are part of everyday work now, and we think that's a good thing. Google has said plainly that it rewards helpful content however it was made. So the question isn't whether you used AI. The question is whether the finished piece is worth someone's time.
Most AI drafts aren't, at least not yet. They read like every other page on the topic: smooth, balanced, general, and easy to skip. Readers leave after two paragraphs. Search engines and AI assistants find nothing new to rank or quote. Content Trace exists to close that gap.
It is not built to catch anyone. It shows you, signal by signal, what makes a draft read as generic, then helps you fix it for the three audiences that decide whether content works: readers, search engines and AI answer engines. Every change is shown before and after, so the final call stays yours.
Anyone can try it for free each month, with no account. Pro and one-time Word Packs are there for people who write and optimize longer pieces, or do it often.
How it works
Content Trace reads a draft across 8 sections and 32 individual signals, from sentence rhythm and word choice to reasoning patterns and emotional texture. These are the same things that separate writing people finish from writing people skim. Each signal is explained in plain language, so you can see exactly what to improve.
Most signals are read by Claude, Anthropic's AI model, which reviews the writing the way a trained editor would. A few are measured directly from the text, such as how much sentence length varies. The analysis settings are fixed, so the same text gets the same or a very close score each time.
Then optimize it: three goals
After an analysis, you pick a goal. The optimizer rewrites the weak spots the analysis found, then scores the new version with the same engine. You compare the two versions side by side, with the Human Score, reading ease and Search & AI-answer readiness before and after. You can run all three goals on the same text.
The optimizer is an editor, not an author. It keeps your facts, numbers, names and quotes. It does not add opinions, feelings or stories you did not write. A fact check compares the new version with your original and flags any name, number or quote that changed. Where a real example or source would make the text stronger, it adds a marker like [Add: a real example from your work] for you to fill in.
Because we do not invent details to game the score, the Human Score may move only a little. Reading ease and readiness usually improve most. No tool can guarantee rankings or AI citations, so review every change before you publish.
Scoring adjusts to the type of content
A white paper and a personal blog post should not be judged by the same rules. Good corporate writing is formal and rarely tells personal stories. A good personal essay often does. A detector that ignores this marks careful professional writing as "AI" too often.
So Content Trace first identifies the type of content, or uses the type you choose. Then it adjusts the analysis:
- Signals that do not fit the type are left out of the score. For example, personal anecdotes do not count for a company blog or a technical document.
- The sections that matter most for that type get more weight.
- The final score is calibrated against real human and AI writing of the same kind, so a typical human-written white paper and a typical human-written essay both land in the human range.
Every report says which type was used and which signals were left out, so you always know how your text was scored.
The Human Score: a diagnostic, not a verdict
The Human Score shows how strongly a draft carries the signals of writing a person thought through. It does not label anyone's work with false confidence. It is a score out of 100. Higher scores indicate stronger human writing signals. Lower scores indicate patterns more consistent with AI generation. Each report also shows a confidence level, which is lower for short texts.
These are probabilistic signals, not verdicts. A score of 72 doesn't mean a text is definitely human-written. It means it shows more human than AI writing characteristics for its type of content.
One limit we want to be open about: AI text that a person has heavily rewritten, or that was prompted to imitate a casual human style, can score in the human range. No detector catches all of it, and we say so in every report.
Who it's for
Content Trace is for anyone who writes with AI and wants the result to be better than the draft:
- Content & SEO teamsTurn AI drafts into posts that rank, get quoted in AI answers and keep readers on the page
- MarketersMake AI first drafts sound like your brand, not like everyone else's
- Writers & creatorsSee where your own draft goes generic and tighten it before you publish
- Editors & publishersGive writers specific, signal-level feedback instead of "this feels off"
- Teachers & academicsA supplementary signal when reviewing work, never the only evidence
Who we are
Content Trace is built and maintained by Colin H at Web Thrive. He has spent 20+ years in B2B SaaS marketing, running content, SEO and AEO programs, and now builds them around AI-assisted content production. Content Trace is the tool he wanted for that work.
The principles behind the product are in our manifesto. Have a question or a feature request? We'd love to hear from you. Get in touch →