Unslop offers a $19 editor for restoring voice to AI-polished drafts

Why it matters

AI editing often standardizes the prose it is meant to improve. Unslop is testing whether preserving a writer's voice can support a focused paid product and API.

Software engineer Spartak is selling Unslop as an editor for writers who use AI to fix grammar, then discover that the machine has flattened their voice. The web service analyzes cadence, structure and word choice, returns a rewrite with a visual diff, and charges $19 a month for its highest tier and API access.

Unslop has been operating publicly since at least June 10, the effective date of its terms and privacy policy. The service's terms describe it as the work of an individual developer.

Spartak says he has spent 15 years building startups, though he does not name them on Unslop's public pages. His reason for building this product is personal. English is not his first language, and he had begun using AI to catch typos, correct small mistakes and find words that did not come readily to him.

The edits worked mechanically. They also gave his writing a recognizable machine rhythm.

"However I used it, the text came back sounding like AI," Spartak writes on Unslop's About page. He describes seeing the same filler, turns of phrase and heavy punctuation across company blogs and social posts. Unslop is his attempt to keep the useful part of AI editing without making every writer sound as though they started from the same prompt.

Editing the edit

Unslop occupies a narrow point in the writing workflow. It is designed for text that has already passed through an AI model, including drafts polished for grammar or vocabulary. The engine then rewrites the result toward what Unslop considers a less machine-typical cadence.

Users receive a before-and-after score and a diff marking insertions and deletions. Unslop says a fidelity check is intended to protect names, numbers and claims while keeping the rewritten draft close to its original length. That claim has not been independently benchmarked, and Unslop's own examples show why fidelity remains a difficult problem.

In one newsletter example, Unslop replaces "embracing this mindset" with "enjoying that mindset." The second phrase is less formulaic, though it also carries a different meaning. That is the central product challenge: generic language can be easy to identify, while deciding which awkward phrases belong to the writer requires a deeper model of voice.

Unslop deliberately avoids promising a score of zero. Its explanation is sensible: human writers sometimes use words associated with AI prose, and removing every such phrase could produce another artificial house style. The engine leaves language alone when an edit might bend the writer's meaning, according to Unslop.

The homepage's live counters on August 16 showed 26 writers, 31 rewrites and about 28,000 characters processed, with a claimed median of 44% "slop removed." Those are self-reported product counters rather than audited usage figures. The percentage is an internal score, not an external measure of quality or authorship. The numbers describe an early product whose immediate job is proving that writers will return after the five free rewrites.

A solo product with an API wedge

The pricing structure starts with five free rewrites of up to 1,000 characters each. A $4.99 monthly plan provides 250 rewrites and raises the per-draft limit to 2,000 characters. The $19 Pro plan includes 600 monthly rewrites, drafts of up to 20,000 characters and API access. Annual prices are $49 and $190, respectively.

The API gives Unslop a route beyond individual writers. Developers can send drafts to the same engine used by the web app, preserve Markdown and retrieve rewritten text through an HTTP endpoint.

That quota favors measured editing workflows over bulk content generation. A newsletter platform, publishing tool or internal communications app could insert Unslop after its existing AI drafting step. The API also places the product closer to WriteHuman's MCP server, which brings humanization or detection into compatible tools such as Claude, Cursor and Codex, rather than requiring users to copy text into a separate website.

Unslop's privacy policy says drafts and rewrites are retained for no longer than 48 hours and excluded from model training. Paid-tier text is sent to Anthropic's Claude API for processing. That dependency keeps the initial engineering burden low for a solo founder, while making Unslop's defensible work the editing process, scoring system, fidelity checks and eventual understanding of each customer's voice.

A different pitch in a difficult category

Unslop enters a crowded group of services commonly called AI humanizers. Many products in that category advertise their ability to evade AI detectors, particularly for student work and search-focused content. Spartak is positioning Unslop around authorship instead: the writer created the ideas, used AI for assistance and wants the finished prose to retain an individual voice.

The technical boundary is porous. Rewriting machine-associated phrases can also affect detector scores, even when detector evasion is absent from the marketing. A 2025 study of 19 humanizers and paraphrasing tools found that many detectors performed worse on rewritten text. The researchers also found that humanizers could reduce fluency, introduce nonsensical phrases or distort meaning, with quality varying widely across products.

An August 6, 2026 preprint reported that two commercial detectors flagged 64% to 80% of lightly AI-refined academic abstracts in its experiment. The same study warned against using detector scores as standalone evidence of misconduct.

That creates a real customer problem without resolving the ethical tension around humanization tools. Unslop can serve writers protecting original work from a generic editing pass. The same engine could be applied to a fully generated draft. Unslop's acceptable-use rules prohibit unlawful activity and spam at scale, while leaving responsibility for submitted text with the user.

Spartak's immediate bet is narrower than the category's most aggressive claims. He is asking writers to pay for a second editing pass that removes the stylistic residue left by the first one. The low price and five free rewrites make that thesis inexpensive to test. The harder work begins after signup: proving that Unslop can recognize a writer's voice, preserve it consistently and avoid replacing one machine mannerism with another.