Hidden Ask: a browser extension that shows text written for AI, not for you

The extension side already has competitors. Four or more small Chrome extensions (Remora, InjectGuard, AI Gremlin Detector, ClawSentinel Guard) detect and some reveal hidden AI-directed text. They are all framed as protection for AI agents, have tiny user bases and are easy to copy, so a stand-alone badge-and-highlight extension would struggle to charge money. It works better as a free funnel. The monitor side is more open. Existing crawler comparison tools are one-off Apify actors, mostly aimed at SEO, and AASA is a security scanner. None offer scheduled crawls of a domain list using both browser and verified AI-bot identities, time-stamped and archived proof of what each crawler saw, change history, or reports written for journalists, consumer groups and brand or GEO agencies. That evidence-and-accountability angle, plus possibly a public index of offenders, is the opening. One risk: early data from Crawl Integrity Verifier suggests AI crawlers are mostly blocked or served the same content as people rather than tailored content. Research also finds most injections are educational or unsophisticated. So paying demand may depend on whether serving AI bots different content becomes a common commercial practice, and free academic and Google research already covers part of the reporting need.

• Remora (Chrome extension)

  Scans each page for 8 hidden-instruction techniques (invisible text, off-screen elements, HTML comments) and shows a blocking overlay. Built to protect AI agents, not to report findings.

• InjectGuard (Chrome extension, MIT open source)

  Flags visually hidden text that matches injection phrases such as 'ignore previous instructions', using display:none, opacity and off-screen checks. Runs locally and has a handful of users.

• AI Gremlin Detector (Chrome extension)

  Closest to the Hidden Ask extension idea: reveals hidden text, normalizes tiny fonts and highlights what it changed, based on how the page renders.

• ClawSentinel Guard (Chrome extension)

  Gives each page a green, yellow or red rating before you share it with an AI agent. Scanning is local.

• Crawl Integrity Verifier (Apify actor)

  Compares what browsers and AI crawlers such as GPTBot and ClaudeBot are served, and outputs a divergence score, a verdict and the changed text. On-demand only; no ongoing monitoring or evidence archive.

• Cloaking Detector (Apify actor)

  Compares Chrome, Googlebot and Bingbot responses and gives a similarity score. Aimed at SEO and does not include AI crawlers.

• AASA - AI Agent Attack Surface Analyzer (open source)

  Crawls sites and PDFs for hidden text and comments aimed at AI agents, using rules plus an LLM judge. A security tool, not a monitoring service.

• Buzur (open source library)

  Python and JS scanner that protects agent frameworks from indirect prompt injection. Built for developers, not a product for reporters.

• Indirect Prompt Injection in the Wild (Khodayari et al., April 2026)

  Academic measurement of hidden instructions on the web: about 70% sit in non-rendered HTML, and most of the HTML-based ones are hidden with CSS tricks. Shows demand, but it is a free public reference.

• Google Security blog: prompt injections on the web

  Google scanned Common Crawl for prompt injections. Most hits were educational or unsophisticated. Big-company research rather than a product.

Checked 2026-10-20