HomeAI Detector › For Blog Comments

AI detector for blog comments, built for community managers.

Screen AI-generated comments, fake-engagement praise, and link-stuffed spam clusters before they reach your readers. Batch scan the pending queue on WordPress, Disqus, Discourse, Webflow, and Ghost. Sentence-level highlights, REST API on Business. Free to try, no card. Your first scan in about six seconds.

Start free, no card See pricing
3 scans/day free WordPress plugin + Chrome extension REST API on Business
Built for community operators

For blog owners, forum mods, and community managers.

A screening tool tuned to the shape of AI-generated comments, not a verdict. The moderator still calls the shot. The detector does the reading.

Every popular post now pulls a wave of polite-sounding comments that say nothing about the article and quietly drop a link to a thin SEO site. The shape is consistent. Three sentences, generic praise, a request to learn more, a URL that does not match the comment. Site owners and community managers see this pattern dozens of times per popular post, and the classic spam filters do not catch it because the prose reads clean.

Blog owners running WordPress, Ghost, or Substack

The pending queue arrives every morning. A solo blogger might see twenty pending comments per day, a popular site sees a few hundred. The free tier covers the solo case. Pro at 19.99 USD per month covers the active blog. The WordPress plugin scans inline, the web app handles Ghost and Substack as a batch paste, and the Chrome extension reads any admin panel that exposes the comment text in a textarea.

Forum moderators on Discourse, Disqus, and Reddit-style stacks

Forum mods see comment clusters fastest. A single operator pushes the same model through five fresh accounts and drops a comment on each of five different posts in an afternoon. Run the batch through the detector and the cluster lights up because every comment scores in the same low band. Pair the Authenticity Score with the new-account flag and an unfamiliar IP range to neutralise the operator inside one round.

Community managers running multi-site publications

Multi-site operations on Business at 39.99 USD per month unlock the REST API and bulk scanning for moderation pipelines. Pipe every new comment through the API as part of the approval webhook, attach the score to the moderation record, and let the human focus on the borderline band instead of reading every polite-but-empty three-sentence reply.

Pattern recognition

The looks-AI comment patterns you see weekly.

Four shapes that cover most of what arrives in a moderation queue. Once you see the rhythm you cannot unsee it, and a scan turns the intuition into a number you can sort by.

1. Generic praise without reference

The most reliable signal. A genuine comment quotes or paraphrases something from the article. The AI praise floats free. Phrases like Great article, I really enjoyed reading this, Thanks for sharing this valuable insight. If the comment could attach to any post on any blog, it probably did.

2. Off-topic generic compliments with a bridge to a link

The comment compliments the article and then offers a related URL the reader might find useful. Your post is about garden design and the linked URL goes to a payday loan comparison page. The whole comment exists to host that link, and the praise is the disguise.

3. Single-link drops on freshly registered accounts

An AI comment cluster almost always rides on freshly registered Disqus, WordPress, or email-only accounts with no comment history. One contextual link, no obvious spam markers, but the account never posted before. Combine the low Authenticity Score with the new-account flag and the false-positive risk drops to near zero.

4. The three-sentence rhythm

Praise, vague engagement, link bridge. Real comments are uneven in length. AI comments cluster around the three-sentence shape because the prompt asks for a polite short reply. Sort the queue by length and the AI tier surfaces.

Plans & pricing

Pick the plan that fits your moderation load.

Solo blogs and small Substacks live inside the free tier. Active publications move to Pro. Multi-site operators and forum networks run the REST API on Business. Full details on the pricing page.

Free
$0/forever

 

Try the detector on your pending queue. No card, no email.
  • 3 scans / day
  • 5,000 chars per scan
  • Sentence-level highlights
  • Chrome extension
Start free
Starter
$7.49/month

Billed $89.88/year — Save $30

For hobby blogs and small Substacks moderating a light queue.
  • 20 scans / day
  • 20,000 AI rewriter words/mo
  • Chrome extension
  • Email support
Get Starter
Business
$29.99/month

Billed $359.88/year — Save $120

For multi-site publishers and forum operators with API pipelines.
  • 100,000 AI rewriter words/mo
  • REST API access
  • 5 team seats
  • Bulk queue scanning
Get Business

Yearly billing saves 25%. View full pricing →

Platform integration

How TextSight fits beside your existing stack.

Akismet and AntiSpam Bee still do useful work on classic spam. TextSight sits next to them and catches what they miss, which is the AI shape itself.

WordPress with Akismet or AntiSpam Bee

The TextSight WordPress plugin installs alongside Akismet or AntiSpam Bee, not instead of them. Classic spam still hits the classic filter and disappears. Everything that makes it past Akismet lands in the pending queue with an Authenticity Score attached. The moderator sorts by score, trashes the bottom tier in one pass, approves the comments that reference a specific paragraph of the article, and reads the middle band by hand.

Disqus and the Chrome extension

Disqus does not expose a clean batch API for moderation, so the Chrome extension does the work. Open the Disqus moderation panel, the extension reads the comment text from the textareas on the page, and the scan returns inline. Same workflow for any admin panel that renders the comment as plain text.

Discourse, Webflow, and Ghost

Discourse forum operators on Business tier pipe new posts through the REST API as part of the approval webhook. Webflow and Ghost moderators paste the pending queue into the TextSight web app as a batch and scan the whole list in under 30 seconds. Per-comment scores and sentence-level highlights arrive together so the moderator can sort and decide in one screen.

REST API on Business for bulk scanning

The Business tier at 39.99 USD per month opens REST API access. Send a POST per comment, get the Authenticity Score and highlight ranges back, attach the result to your moderation record. Most publishers wire this into the approval webhook so the score is in the review queue before a human ever opens the comment.

Workflow

Per-comment vs batch scanning workflow.

Two modes that cover almost every moderation queue. Pick the one that matches the volume.

Per-comment scanning

Right for the WordPress plugin and the Chrome extension. Open a single pending comment, click scan, get an Authenticity Score and a per-sentence colour map back in a few seconds. Useful when the queue is small or when a single comment looks suspicious and the moderator wants a closer read before deciding.

Batch scanning

Right for the web app and the REST API. Paste a queue of 40 pending comments into the textarea or POST them through the API and get a list back with one score and one highlight range per comment. Sort the list by score, trash the bottom tier in one pass, approve anything above 75 that references a specific paragraph, and read the middle band by hand.

The middle band is where the moderator earns their pay

A score in the 50 to 70 band is genuinely ambiguous. Polite real readers land there sometimes. So do mid-quality spam attempts. The detector tells the moderator where to spend the careful reading time and lets the unambiguous tiers process automatically. Two to three minutes a day on a normal blog, ten minutes on a launch day. Far less than reading every polite-sounding nothing one comment at a time.

Cluster detection without configuration

When five comments arrive on five different posts in one afternoon and every one scores in the same low band, that is a cluster. No configuration required. Trash the batch, blocklist the destination domains, and watch the next operator wave hit the wall before it lands in your queue.

Honest scope

A screening tool, not a verdict.

Three things the detector does well and one thing it cannot do. Read both lists before wiring this into a moderation pipeline.

What the detector does well

Surface the AI shape in seconds across a queue of 40, light up clusters with low scores that share phrasing, and reduce a 20-minute moderation block to a two-minute pass. Sentence-level highlights point a moderator at the specific phrases driving the score, so the call is not a black box.

What the detector cannot do alone

Return a verdict on a single short comment with high confidence. A polite four-sentence reply written by a real reader can land in the AI zone because the shape is similar. The moderator still has to weigh the linked URL, the account age, the comment history, and the relevance to the article before deciding. The Authenticity Score is one input among several, not the only one.

The moderator stays in the loop

Auto-trashing a comment because the score is low is the wrong workflow. Auto-holding the comment for human review when the score is low and the account is new is the right one. The detector compresses the moderator's reading load. It does not replace the moderator's judgement, and pretending otherwise eats your community trust faster than the AI spam does.

Community trust is the real prize

Readers leave comments when the visible section feels real. Stack of AI praise, the real commenters stop. Clean moderation, the real conversation returns. The detector exists to keep the comment section a place where the actual readers want to be, and that means accepting the screening output as a hint instead of a verdict.

FAQ

Site owners and moderators frequently ask.

Is TextSight a verdict on whether a comment is AI?
No. TextSight is a screening tool, not a verdict. It returns an Authenticity Score and sentence-level highlights so a moderator can sort a queue by AI-likelihood and triage faster. Moderators should still review the linked URL, the account age, and the on-topic relevance before deciding to trash, hold, or approve.
Which platforms does TextSight integrate with for comment scanning?
The TextSight WordPress plugin sits alongside Akismet or AntiSpam Bee and scans pending comments inline. Disqus, Discourse, Webflow, and Ghost are handled by pasting batches into the web app or by calling the REST API on the Business tier. The Chrome extension also reads comment text inside any admin panel that exposes the comment as a textarea.
What does an AI-generated comment usually look like?
Three patterns dominate. Generic praise that references nothing specific in the post. A tight three-sentence shape with phrases like Great article or Thanks for sharing this valuable insight. And a single link drop where the URL has no relationship to the comment text. A comment that compliments without quoting and links without context is almost always machine-drafted spam.
Do I need the Business tier with REST API access?
Solo bloggers and small Substacks usually sit inside the free tier or Pro at 19.99 USD per month, or 14.99 USD on annual billing. Multi-site publishers and forum operators running automated moderation pipelines move to Business at 39.99 USD per month, or 29.99 USD on annual billing, which unlocks REST API access for bulk scanning the pending queue as part of a comment-approval webhook.
Will the scan ever flag a real human comment as AI?
Occasionally. A short polite reply written by a real reader can land in the AI zone because the shape matches. Use the score as one input among several. Pair the Authenticity Score with the account history, the linked URL, and whether the comment quotes a specific paragraph. The combination catches the AI cluster while protecting the polite but real readers.
How big a queue can a moderator process per pass?
A queue of 40 pending comments scans in under 30 seconds on the web app. A WordPress site running the plugin sees results inline as the moderator opens the pending screen. Forum operators on Discourse can pipe new posts through the REST API on Business and have the score arrive in the review queue automatically, before a human ever opens the post.
How does this differ from Akismet or AntiSpam Bee?
Akismet and AntiSpam Bee catch classic spam. Link-heavy, broken grammar, named drug brands. The new AI shape is grammatically perfect and often carries a single contextual URL, so the classic filter waves it through. TextSight reads the actual prose and flags the AI shape regardless of whether the comment carries a link, so it sits beside Akismet rather than replacing it.
What happens to the comment text after scanning?
Comments are scanned in memory and the response carries the score and highlight ranges only. Free and Pro scans are not stored beyond the session unless the moderator saves them to scan history. Business tier scans through the REST API can be configured to drop the comment payload as soon as the score returns, which keeps moderation pipelines clean for sites with strict community privacy policies.
Related

More guides for publishers and moderators.

Clean comments. Real readers.

Free to try. No card. Sentence-level highlights and an Authenticity Score for every comment in your queue.

Start free, no card See pricing
WordPress plugin · Chrome extension · REST API on Business · Sentence-level highlights