Home › FAQ

Frequently asked.

Search 30+ answered questions, or filter by category. Can't find what you need? Email us — we reply fast.

Detection & accuracy

How accurate is the TextSight AI detector?
99.2% on our internal benchmark — 1,000 documents mixed across GPT-4o, Claude 3.5, Gemini 1.5, Llama 3, and human writers. False-positive rate is 0.8%. Read the full methodology →
What AI models do you detect?
ChatGPT (GPT-4o, GPT-4, GPT-3.5), Claude 3.5 (Sonnet, Opus, Haiku), Gemini 1.5 (Pro, Flash), Llama 3, Mistral, DeepSeek, Perplexity, and dozens more. We retrain with every major model release.
What's the Authenticity Score?
A 0–100 number that tells you how human your writing reads. We combine perplexity, burstiness, stylometry, and embedding signals — calibrated against the detectors that grade your work (Originality, GPTZero, Copyleaks, Winston, Turnitin). Higher = more human.
Will it flag my human writing as AI?
It's the thing we obsess over. We trained on 4.2 million human samples — including formal, academic, and "AI-sounding" human writing — to keep false positives below 0.8%. If you do get a wrong flag, every paid plan includes free re-review by a human grader within 24 hours.
What languages does TextSight support?
The AI Rewriter and writing tools work in 50+ languages — English, Spanish, French, German, Portuguese, Italian, Dutch, Polish, Russian, Arabic, Hindi, Chinese (Simplified & Traditional), Japanese, Korean, Turkish, Vietnamese, Indonesian, and more (the top 18 fully supported, the rest in beta). AI detection is optimized for English — other languages still return a score, but accuracy is lower.
Can it detect AI-rewritten or paraphrased text?
Yes. Vocabulary swap and paraphrase passes leave stylistic fingerprints in burstiness and perplexity — we trained specifically against these patterns. We also detect text rewritten by major AI rewriter tools.
How does sentence-level highlighting work?
Every sentence gets its own AI probability score — color-coded red (high), amber (mixed), green (human) right inside your text. Hover any sentence to see its per-classifier vote and the suggested rewrite.
What's the longest text I can scan?
Free: 5,000 characters. Starter: 10,000. Pro: 25,000. Business: 50,000. Enterprise: 100,000. For larger documents, the bulk scanner splits them automatically.
Do you offer a sample report?
Yes. Browse a full sample report → — sentence-level scores, model attribution, and the Authenticity Score, all on one page.

AI Rewriter & rewriting

Does the AI Rewriter perform well across other AI quality checkers?
Yes. We continuously test against Originality, GPTZero, Copyleaks, Winston, and Turnitin. On the same 1,000-doc set, rewritten text scores under 5% AI on all five — facts and meaning preserved.
Will the AI Rewriter change my facts?
No. Numbers, names, dates, and citations are locked as anchors before rewriting — we treat them as fixed values and only edit prose around them.
What are the tone presets?
Conversational, Professional, Academic, Blog post, Email. Business and Enterprise plans get custom tone training — feed it your own samples.
How are Rewrite Suggestions different from the AI Rewriter?
Rewrite Suggestions are per-sentence proposals you can accept or reject one at a time with a diff view. The AI Rewriter rewrites the whole text in one shot. Use Suggestions for surgical edits, AI Rewriter when you want to start fresh.
How many AI rewriter words do I get?
Free: none. Starter: 2,000/mo. Pro: 10,000/mo. Business: 40,000/mo. Enterprise: 150,000/mo. Custom volumes available — contact sales.

Pricing & billing

Is the free plan really free, forever?
Yes — forever. 3 scans a day, up to 5,000 characters each, plus full access to all 20 free writing tools. No card. No trial expiration. Upgrade only when you need volume.
How does the student discount work?
Pro plan is $13.99/mo (instead of $19.99) with a verified .edu email. The discount renews as long as the .edu stays active.
Yearly billing — is it really 20% off?
Yes — toggle Yearly on pricing. We bill once a year; you save effectively 2 months. Cancel anytime, refund of unused months.
Do you offer refunds?
Yes. Cancel within 7 days of any paid renewal for a full refund. After that, we pro-rate the remaining days back to your card. No questions, no awkward forms.
What happens if I exceed my limits?
We don't auto-charge overages. You'll see a friendly nudge to upgrade or wait for the next billing cycle. Scans pause when you hit the daily cap; characters truncate with a warning.
How do I cancel?
Settings → Billing → Cancel. One click. We pro-rate and refund unused days. No contracts.
What payment methods do you accept?
All major cards (Visa, Mastercard, Amex), Apple Pay, Google Pay, and bank transfer for Enterprise. Powered by Stripe — your card data never touches our servers.

Privacy & security

Is my text safe?
Yes. Your text is processed in memory and discarded the moment we return your result. Zero retention by default. Never used to train our models. Never seen by humans. Read the content-handling policy →
Are you SOC 2 certified?
SOC 2 Type II audit is in progress (expected Q3 2026). We're GDPR-compliant today, with a DPA available on request for Business and Enterprise plans.
Do you store my scan history?
Only if you opt in. On paid plans, "Save to history" is opt-in per scan — encrypted at rest with AES-256, region-pinned (US or EU), deletable in one click.
Are you GDPR-compliant?
Yes. EU data stays in EU regions. Right to deletion, export, and rectification supported. Sign a DPA via sales.
What about the Chrome extension's data?
The extension only sees what you actively select. No background browsing data, no cookies, no analytics on pages you visit. Full extension privacy →

Account

Do I need an account to try the detector?
No. The detector and most tools run free for anonymous users with daily limits. Sign up only to save history, sync devices, or upgrade to a paid plan.
Can I sign in with Google?
Yes — Google, Microsoft, and email all supported. SSO (SAML, Okta, Azure AD) on Enterprise.
How do I switch from monthly to yearly billing?
Settings → Billing → Change plan. We pro-rate the remaining month against the yearly upgrade.
Can I add team members?
Yes — Business (5 seats) and Enterprise (10 seats). Invite by email; everyone gets their own login, shared workspace, and admin controls.

API & integrations

Do you have an API?
Yes — REST API on Business plan and above. 10K calls/month on Business, 50K on Enterprise. Read the docs →
What's the API latency?
p95 latency 380ms for a 5,000-character scan. The bulk endpoint streams sentence-level results as they finish, so you get progress feedback.
Are there official SDKs?
Python, Node.js, PHP, Ruby, and Go — all open source on GitHub. See the docs →
What integrations are available?
Google Docs, Microsoft Word, Notion, WordPress, Shopify, Canvas LMS, Slack, Zapier. Custom integrations for Enterprise via the API.
Do you have webhooks?
Yes, on Business and Enterprise. Subscribe to scan-completed, scan-failed, and AI rewriter-completed events.

No matches. Try a different keyword, or email us directly.

Deep dive

Power-user questions.

A second pass of detail-level questions we get from agencies, integrators, and editorial teams after they have used TextSight for a few weeks. These answers go beyond the basics — pick the ones that fit your workflow.

How accurate is TextSight on long-form versus short-form content?
On our internal benchmark we measure 88 to 92% accuracy on long-form passages above 500 words, where the model has plenty of stylistic signal to read — burstiness, vocabulary spread, sentence-rhythm variance, and tonal shift across paragraphs all contribute. Short-form content under 150 words sits at 70 to 78% accuracy because there is simply less surface area to evaluate. We publish the full methodology and dataset composition on the accuracy methodology page so you can judge the numbers in context. Treat any single-sentence verdict as a hint, not a conclusion, and rely on the sentence-level highlights for longer passages.
What is the difference between Authenticity Score and the AI percentage used by competitors?
A competitor AI percentage answers a single yes-or-no question: how likely is this whole document AI-generated. The Authenticity Score answers a more useful question: how human does this text read on a structured 0–100 scale, broken into five named bands — Original, Mostly Human, Mixed, Likely AI, AI Generated. The score factors in vocabulary spread, sentence burstiness, and tonal variance alongside raw probability. Writers find the band labels easier to act on than a bare percentage, because the bands map directly to the rewrite intensity needed.
How often is the detection model retrained?
Roughly every six to ten weeks for incremental retraining cycles, and immediately whenever a major frontier model ships (GPT-5, Claude 4, Gemini 2, Llama 4) so detection coverage does not drift. Each retraining cycle is logged on the methodology page with a short note on what changed and which model family it improves coverage for. You do not need to do anything on your end — the live API always points at the latest production model, and your saved scans are never overwritten by re-scoring.
How does the closed-loop AI Rewriter actually work under the hood?
It is a generate-then-score loop. The first rewrite pass produces three candidate versions tuned to the tone you picked. Each candidate is run back through the detector. If the highest-scoring candidate still reads above 35% AI on the closed-loop threshold, the loop iterates — generating new variants seeded by the parts that scored lowest — until the result is consistently under 35% across all five reference detectors (Originality, GPTZero, Copyleaks, Winston, Turnitin). Most documents converge in two passes; complex academic prose can take three. You see the result stream in token-by-token through SSE so there is no spinner-and-wait UX.
Can the Chrome extension work in incognito mode?
Yes, but you have to enable it explicitly. Open chrome://extensions, find TextSight, click Details, and toggle "Allow in incognito" to on. After that the right-click scan action and inline Authenticity Score work in incognito tabs exactly as they do in normal windows. Scan history is still tied to your TextSight account, so anything you scan while signed in appears in your history regardless of browser mode. If you want zero-history scans, sign out before opening incognito.
How do I integrate TextSight with my CMS?
Three paths depending on your stack. For WordPress, install the official TextSight plugin and connect it from /wordpress-connect — scans run from inside the post editor. For Webflow, Ghost, Sanity, Contentful, or any headless CMS, use the REST API on the Business plan to wire detection into your editorial workflow. For Notion-based workflows, the Chrome extension covers most ad-hoc needs without setup. Full API specs and example payloads are on the API docs page, including curl snippets and Python/Node/PHP/Ruby/Go SDK samples.
What happens to my scan history if I downgrade?
Existing scans remain visible during a 14-day grace period after the downgrade so you can export anything important. After the grace window closes, scans older than your new plan's retention window are removed. For example, dropping from Pro (30 days) to Starter (14 days) means scans aged 15 to 30 days disappear once grace ends. Account-level export still works during the grace window, and a downgrade never deletes your account, your invoices, or your API keys.
Can I export my full scan history?
Yes. Go to Settings → Data Export and request a JSON bundle of every scan, including timestamps, input text, Authenticity Score, AI percentage, sentence-level breakdown, and any rewritten output you generated. The bundle arrives by email within 15 minutes for accounts under 10,000 scans, and within a few hours for larger archives. Business plan users can also pull the same data programmatically through the REST API, paginated 1,000 scans per request.
Pick a plan

Match a plan to your scan volume.

Four tiers, no overage charges, one-click upgrades. Most newcomers start Free, then move to Pro after the second week of regular use.

Free
$0

Best for: Evaluating the product, students checking a single essay, or writers running an occasional spot-check before publishing.

3 scans per day, 5,000-character scan ceiling, two lifetime AI rewriter uses, signup not required for the first scan.

Starter
$9.99/mo

Best for: Active students with three to five graded essays per week, hobby bloggers, freelancers checking drafts before client delivery.

20 scans/day, 20,000 AI rewriter words/month, Chrome extension, plagiarism risk signal, 14-day scan history.

Most popular
Pro
$19.99/mo

Best for: Daily newsletter writers, SEO writers shipping 10+ pieces per week, single-seat freelancers running detection across client work.

Unlimited scans, 50,000 AI rewriter words/month, file + URL upload, priority support, 30-day scan history.

Business
$39.99/mo

Best for: Teachers grading whole-class submissions, agencies with 3+ writers, SaaS/EdTech teams embedding detection through the REST API.

100,000 AI rewriter words/month, REST API access, five team seats, white-label PDF reports, 90-day history.

Annual billing saves 25% — Pro drops to $14.99/mo, Business to $29.99/mo. Full pricing breakdown →

Still stuck?

Real humans, fast replies. Pro & Business get answers within 4 business hours.

Contact us Try TextSight