StealthWriter AI Review

I did not want to review this tool from its marketing page, so I ran my own text through it on the free plan and checked every result against an outside detector. Here is what happened, output by output, with the exact scores I saw. Everything below was produced on the free tier using the Ghost 5.2 Mini model, which is the only model free users get.

Test 1: A casual blog intro

I started with a 140-word blog intro about meal prepping, written in ChatGPT in a loose, conversational tone. Pasted straight into StealthWriter, its own V2 scanner rated it 0% human and tagged it as AI. That is the starting line.

I set the level to 8 (heavy rewrite) and hit humanize. The rewrite came back readable. Lines like "you arrive home, fatigued and just wanting to eat" flow fine, and the tool's own V1 scanner jumped the score to 75% human with a green "Looks Human" badge. On its own dashboard, this looks like a pass.

Then I took that exact humanized text and dropped it into GPTZero. The verdict flipped completely. GPTZero was highly confident the text was AI generated and scored it 100% AI, 0% human. The same paragraph StealthWriter called human, an outside detector called machine-written with no hesitation.

Worth flagging that even StealthWriter's two scanners disagree. Its V2 read the original at 0% and its V1 read the rewrite at 75%, so the built-in score shifts depending on which version you run. I would not treat that internal number as proof of anything.

The sentence rewrite feature

One part I did like. Click any sentence and StealthWriter offers alternative rewrites to swap in. On my meal-prep line it gave three options, and it flags each one for AI risk with a colored dot. Option 3 carried a red dot, meaning the tool itself rates that rewrite as high AI impact. So the feature gives you control, but it also quietly admits some of its own suggestions are worse than others.

Test 2: A technical paragraph with numbers

Next I tested whether facts survive the rewrite. I fed it a 94-word paragraph comparing 4G LTE and 5G, packed with figures: 20 to 100 Mbps, 100 to 900 Mbps, a 1,000 Mbps peak, and the 3.5 GHz band. Two things showed up right away. The V2 scanner again read 0% human, and a banner warned that results may be off because the text sits under 100 words. The same screen shows the free ceiling: 94 of 1,000 words used, plus a "Limited daily scans" note that resets at 5:30 AM.

The humanized version scored 80% human on StealthWriter's V1. The genuinely good news is in the numbers: every figure came through intact. 20 to 100 Mbps, 100 to 900 Mbps, the 1,000 Mbps peak, and the 3.5 GHz band all survived the rewrite unchanged. For a tool that reshuffles sentence structure this heavily, keeping the technical facts accurate is a real point in its favor.

Where it broke: formatting

This one is a practical headache. I built a properly formatted draft on filing taxes early, with a bold H2 heading, three bullet points, one bolded phrase, and a hyperlink to the IRS filing page.

The moment it landed in StealthWriter, all of that was gone. The heading dropped to plain body text, the bullets lost their markers and collapsed into run-together lines, the bold vanished, and the IRS link became flat, unclickable text. StealthWriter reads pasted content as plain text only.

The takeaway for anyone working with finished posts: humanize first, then format. If you paste a styled draft in, you will rebuild every heading, bullet, and link by hand afterward.

What the free plan actually gives you, and what it costs to go up

The free tier is more usable than I expected. It includes the humanizer, both detector versions, deep scan, and a daily allowance of 10 humanizations and 10 scans at up to 1,000 words per input. That is enough to genuinely evaluate the tool, which is why this whole test ran on it.

Paid plans climb steeply. Starter at $20 a month raises the caps to 50 a day and 5,000 words, and it unlocks Ghost 5.2 Pro plus the Legacy models. Plus is $50, Pro is $100, and the top self-serve tier, Scale, runs $400 a month for 2,000 humanizations a day.

The pricing detail that matters for my results: the free plan only runs Ghost 5.2 Mini. Pro and Legacy models are locked behind Starter and up. So my GPTZero failure is specifically a Mini-model result. The paid models might do better against detectors, though I could not verify that on the free tier. On the free plan as tested, StealthWriter reliably fools its own scanner but not GPTZero, and GPTZero is the detector students and freelance clients actually run.