AI Slop Detection

What is it

AI Slop Detection is the technical and commercial practice of identifying machine-generated or machine-polished text — the telltale "AI flavor" of over-regular rhythm, hedge-stacked sentences, listicle scaffolding, em-dash tics, and semantically empty transitions — and either flagging it or rewriting it back into something that reads like a human wrote it. Technically, this splits into two layers. The first is a classifier: a model (often a fine-tuned small transformer or a perplexity/burstiness heuristic) that scores text on a probability of AI origin. The second, and commercially more interesting, is a normalizer: a rewriting pipeline that strips the patterns without destroying meaning.

The business significance is bigger than "detect ChatGPT." Buyers don't actually want a detector — they want to stop getting punished for sounding like a bot. That means SEO teams worried about Google's helpful-content signals, agencies whose clients can smell generated copy, educators, and any founder publishing a blog that currently reads like a LinkedIn thought-leader. Detection is the hook; remediation is the product. The terms surfaced on Show HN and w2solo in late October 2026, with some entrants already shipping OpenAPI and MCP interfaces or running fully in-browser — a signal that the builders see this as infrastructure, not a consumer novelty.

Why now

Three forces converged in 2025–2026 to make this a real category rather than a joke.

First, saturation. By mid-2026 the volume of AI-assisted text on the public web crossed the point where readers and algorithms both started discounting it. Google's spam and helpful-content updates have been progressively hostile to scaled, low-effort generated content — publishers who got hit lost real traffic and now have budget to prove their copy is human.

Second, the models got more detectable in aggregate even as individual outputs improved. The diffusion of a handful of dominant models means their stylistic fingerprints are now well-characterized — burstiness, token-frequency signatures, and structural tics are stable enough to classify reliably. A detector built today has a moving but tractable target.

Third, distribution infrastructure matured. MCP (Model Context Protocol) and OpenAPI-first tooling mean a detection/rewrite service can plug directly into the editor, CMS, or agent workflow where the text is actually produced. That's the difference between a 2010s "plagiarism checker" SaaS and a 2026 in-line middleware.

The timing is narrow, though. If OpenAI, Anthropic, or Google ship a first-party "humanize" mode, the standalone category compresses fast. The window is the next 12–18 months.

Market Evidence

The raw signal is thin and early: 2 independent sources (w2solo and Show HN), 2 total mentions, first seen 2026-10-25, stage "nascent," growth rate 100% (which, off a base of 2, means almost nothing statistically), and a trend score of 66/100. The opportunity, market, competition, and demand scores all read 0/100 — that's not a verdict that there's no market, it's a verdict that the scoring model has essentially no data to work with yet.

Here's the honest read: this is a leading indicator of a real problem, not yet proof of a paid market. Show HN is where builders post things they're personally annoyed by — and "my AI-written copy smells like AI" is a genuine, widespread annoyance. But annoyance ≠ willingness to pay, and two mentions is a whisper, not a wave.

What makes it worth watching rather than dismissing: the entrants already offer OpenAPI and MCP interfaces and in-browser execution. Builders don't invest in MCP plumbing for a joke project. That's a signal of intent to be infrastructure. Treat the 66 trend score as "interesting," the 0 scores as "unmeasured," and go validate demand yourself before writing code.

Who's Behind It

There are no whales yet — that's the point. The drivers are individual indie builders and small teams posting on Show HN and w2solo, likely solo founders who hit the problem in their own content workflow. The competitive dynamics are therefore wide-open but also unstable: the moment this shows revenue, it gets cloned in a weekend, and the moment it shows real revenue, a platform absorbs it.

The entities to watch aren't startups — they're the platforms that could make or break the category. OpenAI, Anthropic, and Google control the models whose output is being detected; any of them shipping a native "human tone" control is an extinction event for standalone detectors. On the demand side, the SEO and content-marketing tooling incumbents (Surfer, Clearscope, Originality.ai) are the natural acquirers or fast-followers. Originality.ai in particular already sells AI-detection to publishers and has the distribution to bolt on a "de-slop" rewrite feature.

Your realistic competitive set for the next year is other indie hackers. Your realistic existential threat is a platform feature. Build accordingly: move fast, own a niche, and stay acquirable.

TAM & Market Size

The buyers are content teams, SEO agencies, indie publishers, educators, and developers who need it as an API. Sizing it honestly: the global AI-content-detection market is estimated in the low hundreds of millions (roughly $150M–$400M depending on the analyst) and growing fast, but that figure is dominated by academic-integrity vendors. The commercial content slice — marketing and SEO — is smaller today but has higher willingness to pay per seat because the ROI is directly tied to traffic and revenue.

Price tolerance: agencies already pay $99–$500/month for SEO suites and $30–$100/month for individual content tools. A "de-slop + humanize" tool that sits in the publishing workflow can credibly charge $29–$99/month per seat, or $0.002–$0.01 per 1,000 words on an API meter. Developers will pay per-call; agencies will pay per-seat; enterprises will pay for an audit trail.

The demand score of 0/100 reflects missing data, not missing demand — but don't confuse the two. The real question is whether buyers want a detector (cheap, commoditizing, free tools exist) or a fixer (sticky, workflow-embedded, defensible). Bet on the fixer. The detector is the lead magnet.

Competitive Landscape

Current players are fragmented. On detection: Originality.ai (publisher-focused, established, $14.95+ per month), GPTZero (education-heavy brand), and a long tail of free browser tools with poor accuracy. On humanizing/rewriting: Undetectable.ai and a swarm of SEO "AI humanizer" sites, most of which are thin wrappers around a paraphrasing prompt with shaky quality and no real API. Competition score of 0/100 means the field is genuinely uncrowded at the quality-and-infrastructure end — everyone is fighting at the consumer-wrapper end.

The gaps are obvious and exploitable. Nobody credible owns "MCP-native de-slop that plugs into Cursor, Claude, or a CMS." Nobody offers a clean per-word API with a real quality benchmark. Nobody has built the verification layer — proving to a client that copy was de-slopped and scoring it before/after.

If Big Tech enters (a native "human mode" in a major model), you have roughly 6–12 months before the standalone use case erodes. Your defense is workflow integration and a niche (e.g., legal marketing, academic abstracts, non-English markets) that a generic model feature won't serve well. Don't try to out-model OpenAI; out-integrate it.

Business Model

Go freemium SaaS with a metered API tier — hybrid, not one or the other.

Why: the detector is a perfect free hook (run a paragraph, get a score, feel the sting), while the rewrite/normalize step is where the value and the compute cost live. Charge for remediation and volume.

Suggested pricing:

• Free: 1,000 words/month, detection only, browser-based.
• Pro: $29/month — 50,000 words, rewrite + detection, CMS/editor plugin, before/after scoring.
• Agency: $99/month — 250,000 words, 5 seats, client-ready reports, API key.
• API: $0.004 per 1,000 words metered, with volume discounts above 5M words/month.

Rationale: $29 sits below the "expense it without asking" threshold and above the free-tool noise; $99 matches what agencies already pay for a single SEO tool; per-word API pricing aligns cost with the dominant variable cost (inference).

12-month forecast (solo/small team):

• Conservative: 150 paying seats averaging $40 → ~$6K MRR.
• Base: 500 seats averaging $45 + modest API usage → ~$25K MRR.
• Optimistic: 1,500 seats + an agency/API contract or two → ~$75K MRR.

CAC: near-zero via Show HN, SEO on "AI humanizer" terms, and developer communities; paid CAC if you run ads likely $40–$80, payback under 2 months at $29–$99 pricing. Keep it organic-first — the margins don't support heavy paid acquisition at these price points.

MVP Blueprint

Scope hard to 2–7 days. Cut everything that isn't "detect, rewrite, prove."

Core features (only these):

1. Paste-in text box → AI-probability score with a short explanation of which patterns triggered it (burstiness, hedge density, listicle structure).
2. One-click rewrite that removes the flagged patterns while preserving meaning and facts.
3. Before/after side-by-side with both scores.
4. A public REST endpoint (POST /score, POST /humanize) with an API key.
5. An MCP server exposing the same two functions so it plugs into Cursor/Claude.

Tech stack for speed: Next.js on Vercel for the UI and API routes; a small fine-tuned classifier (or a hosted detection API as a stopgap) for scoring; a constrained rewrite prompt against a cheap fast model (e.g., GPT-4o-mini or Claude Haiku class) with a strict "no new facts, no new claims" system prompt; Postgres (Supabase) for keys and usage metering; Stripe for billing.

Fastest path to launch: ship the free detector + MCP server first (day 1–2), gate the rewrite behind a Stripe checkout (day 3–4), add the before/after report and a landing page with real examples (day 5–7). Launch on Show HN with a brutally honest title and a live demo. Do not build accounts, teams, or a dashboard until someone asks.

Commercial Opportunities

1. De-slop API for agent and CMS builders. Sell the detection+rewrite as a metered API and MCP server to developers building content pipelines. Target: indie SaaS founders and agency devs wiring AI into publishing. Expected: $2K–$15K/month at $0.004/1K words once you have a handful of integrations. Beats alternatives because it's infrastructure — recurring, sticky, and doesn't require you to win a consumer brand war.

2. Agency white-label reporting tool. A branded "content authenticity report" agencies hand to clients: before/after scores, pattern breakdown, PDF export. Target: 5–50 person SEO/content agencies. Expected: $3K–$20K/month at $99–$299/seat. Beats alternatives because agencies will pay for proof they can show a client, and nobody sells that artifact today.

3. Niche vertical humanizer. Pick one vertical where tone is regulated and generic models fail — e.g., legal marketing copy, medical-adjacent content, or non-English markets (German, Japanese) where slop patterns differ. Expected: $1K–$8K/month. Beats alternatives because a generic model feature won't tune for a vertical's register, and you own the vocabulary.

Product Ideas

🥇 SlopScore + Humanize (the core SaaS). One-line: "Paste your copy, see how AI it smells, fix it in one click." Target: content marketers, SEO teams, indie founders. Why now: AI-written copy is everywhere and getting penalized; the fix is a 30-second workflow, and no credible tool owns the before/after proof.

🥈 SlopMCP (developer infrastructure). One-line: "The de-slop layer for your AI content pipeline — one MCP server, two endpoints." Target: developers building agents, CMS integrations, and publishing tools. Why now: MCP adoption is accelerating and almost nobody is shipping quality content-normalization as MCP-native infrastructure — first-mover advantage is real for the next year.

🥉 Authenticity Report for agencies. One-line: "Client-ready proof your content reads human — scored, explained, exported." Target: SEO and content agencies with skeptical clients. Why now: agencies are getting asked "is this AI?" and have no artifact to answer with; a branded report is a high-margin, low-churn wedge into the agency segment.

SEO Opportunity

Search demand for "AI detector," "AI humanizer," and "make AI text sound human" is large and growing, but the head terms are brutally competitive (Originality.ai, GPTZero, and content farms own them). SEO difficulty reads 0/100 in the data, which reflects the emerging long-tail, not the head.

Target long-tail instead: "remove AI tone from text," "why does my writing sound like ChatGPT," "AI slop detector API," "MCP humanize text," "de-slop SEO content." Low competition, high intent, developer- and marketer-flavored. Content strategy: publish real before/after teardowns of famous AI-sounding copy — those pages rank and convert because they demonstrate the product.

Risk Assessment

The thesis breaks if buyers only want free detection and won't pay to fix. That's the #1 risk: detection is commoditizing toward free, and the paid rewrite has to be visibly better than a free ChatGPT paraphrase. Validate by charging from day one — a $29 Stripe link and 10 real signups beats 1,000 free users.

Risk #2 (market): a major model ships a native "human tone" control and the standalone use case collapses. Mitigate by owning workflow integration and a vertical a generic feature won't serve.

Risk #3 (execution): quality. A humanizer that changes meaning or introduces errors destroys trust instantly. Benchmark ruthlessly against a held-out set and publish your accuracy.

Cheap validation before building: post three before/after examples on X and Show HN, link a waitlist with a $29 pre-order, and DM 20 agencies asking what they'd pay for a client-ready authenticity report. If you can't get 10 pre-orders or 5 agency conversations in two weeks, walk away.

Action Plan

Today: write a one-paragraph spec of the two endpoints (/score, /humanize), pick your detection approach (hosted API as stopgap is fine), and draft the Show HN post. Set up a Stripe payment link for a $29 Pro tier before you write product code.

Week 1 — validate: build a no-login demo (paste → score → rewrite), ship the MCP server, and post on Show HN and w2solo. Target: 50 demo uses, 10 email captures, 3 pre-orders. Talk to 5 agencies about the report idea.

Month 1 — monetize: add API keys, metering, and the before/after report. Get to 20 paying customers or $1K MRR. Publish two SEO teardown posts.

Month 3 — defend: pick one vertical or the developer-API lane and go deep; add a CMS plugin; benchmark quality publicly. Goal: $5K–$10K MRR and one integration partnership. If MRR is flat and pre-orders stalled, sunset and reuse the pipeline elsewhere.

Related Terms

AI Content Humanization — the rewrite half of this category, already spawning consumer tools; it's the monetizable side of AI Slop Detection.

MCP-native developer tooling — the Model Context Protocol wave that lets detection/rewrite plug into agents and editors; this is the distribution channel that makes the API play viable.

Helpful Content / AI-content SEO penalties — the policy pressure from search engines that gives buyers a financial reason to care, and the demand driver behind agency willingness to pay.