About SignalTools

Why this site exists

Scams, fake content, inaccessible interfaces, and messy data have become daily friction for ordinary people. The answers usually exist, but they are buried inside heavyweight platforms that demand a signup, a subscription, or an email address.

We think a lot of these problems are better solved by small, focused tools that do one thing brilliantly and get out of the way. That is what SignalTools is. Every tool on the site is designed to take a single input — a suspicious message, a screenshot, a CSV — and return a clear, useful answer in seconds.

Who runs it

SignalTools is an independent project built and maintained in the United Kingdom. The project is self-funded and supported by a combination of display advertising (Google AdSense) and affiliate links, which allows the core tools to remain free permanently.

If you want to talk to a human, the Contact page gets straight through.

How we build

What is here today

ScamSupport is the flagship tool. Paste an email, text message, WhatsApp, screenshot, QR code, or even a voice message, and it checks sender authenticity, spots manipulation tactics, analyses links, and returns a plain-English verdict with reasons and a recommended next step. It supports seven input types: email text, URLs, phone numbers, voice messages, images, QR codes, and raw .eml files.

Additional tools are in development and will go live as they meet our quality bar. We would rather ship one solid tool than a half-dozen mediocre ones.

How we make money

Two things: display advertising from Google AdSense, and affiliate links where we genuinely recommend a product (for example, NordVPN). We never base tool results on which partners pay us, and we disclose affiliate relationships where they apply. Full detail is in the privacy policy.

Editorial standards

Scam examples published in the Live Scam Feed and in our scam-type guides are based on current reports, public advisories from organisations such as Action Fraud and the NCSC, and messages submitted by readers. We do not reproduce identifying victim information, and we anonymise specimen messages.

Corrections & takedown policy

SignalTools publishes verdicts about firms on the Investment Scam Platform Lookup, scam-type guides, and the blog. Every verdict cites at least one authoritative source. We take accuracy seriously and we welcome corrections.

If you believe a verdict about a firm you operate is incorrect, email corrections@signaltools.org with:

• The firm name and the page URL of the verdict you're disputing.
• The specific claim you believe is inaccurate.
• Supporting evidence (regulator authorisation, court orders, or other authoritative documentation) showing the claim is wrong or out of date.
• A contact name and verified contact details for the firm.

Our response commitments:

• Acknowledge within 48 hours. Every correction request receives an initial response within two working days.
• Resolve within 5 working days. Where the dispute is substantive, we update or remove the entry within five working days of accepting the correction; where we maintain the verdict, we explain why with citations to the supporting source.
• Public correction log. Material corrections are noted on the affected page with the date of correction and a short note describing what changed.

Requests sent from a non-firm email address or without supporting evidence may be deprioritised. Requests that are part of, or appear to be part of, a coordinated attempt to suppress legitimate consumer-protection reporting will be declined and referred to legal advice as appropriate.

For general feedback or press enquiries (not corrections), use the Contact page or info@signaltools.org.

AI statement & disclaimer

Some tools on SignalTools use machine-learning models to help classify suspicious messages. Where AI is involved, we are explicit about it. This section explains what we do, what we don’t do, and what you should know.

What we use AI for

• Message classification in the Scam Message Scanner. When you paste an email or text into the tool, a small machine-learning model (around 18 KB, runs entirely in your browser) helps score the message against patterns of known scams. The classification is one of several signals the tool produces; final wording is reviewed by humans during template authoring.
• Pattern detection for Business Email Compromise variants (CEO fraud, vendor invoice fraud, payroll diversion). These use rule-based detection with word-list matching, not large language models. The rule set has been validated against a 55-case labelled test set with 100% precision and 100% recall as of May 2026.
• Future use: if we add optical character recognition for uploaded screenshots, or large-language-model-based classification for novel scam patterns, this section will be updated before those features ship.

What AI we use

The current ScamSupport message-checker runs a local browser-only model. This means:

• The message text you paste never leaves your device.
• No cloud AI service receives your message.
• No third-party AI provider (OpenAI, Anthropic, Google, etc.) is involved in classification.

If we ever introduce cloud-based AI (for example, for long-form content analysis), we will name the provider explicitly here, explain the data-handling, and provide an opt-out path.

What we don’t use AI for

• Verdicts on the Investment Scam Platform Lookup. Every “Known fraud”, “Authorised”, “Unauthorised in UK”, “No warning found” and “Possible impersonation” verdict is derived from authoritative regulator sources (FCA Warning List, FCA Firm Checker, SRA register, OCCRP investigation, etc.) and human-reviewed before publication. AI does not assign verdicts.
• Recovery advice. The procedural guidance in our recovery articles (PSR APP claim steps, chargeback timing, Financial Ombudsman escalation) is derived from published regulator rules, UK Finance code documents, and statutory sources — not generated by AI.
• Editorial copy. The brand-impersonation guides, recovery walkthroughs and platform write-ups are human-written and human-reviewed. AI may assist with drafting, but every published page is checked by a human before going live.

Your data and AI

• ScamSupport message-checker: your message stays in your browser. Nothing is uploaded for classification. We retain no message-text logs.
• Tool usage analytics: we may collect anonymous aggregate usage data (e.g. how often each tool is used, country of visitor) via Cloudflare Web Analytics, which doesn’t use cookies or fingerprinting. See the privacy policy for full detail.
• Affiliate clicks: when you click a NordVPN link, the URL contains a UTM tag identifying which page you came from. NordVPN’s own systems track the click as part of the affiliate programme. We never see your NordVPN signup information.

AI limits and disclosure

Machine-learning classification is helpful, but it is not infallible. The Scam Message Scanner can produce false positives (flagging legitimate messages as suspicious) and false negatives (failing to flag genuine scams). The tool’s output is one input to your decision, not the decision itself.

Specifically:

• Validated BEC detector precision and recall: 100% on a 55-case labelled set (May 2026). Production traffic may surface false positives or negatives we haven’t seen in testing.
• The browser-only message classifier was trained on a curated corpus of UK and global scam patterns up to mid-2025. New scam patterns may not be recognised until the model is updated.
• We re-evaluate detection accuracy quarterly. Material accuracy regressions are published in the relevant tool’s release notes.

Human in the loop

The most important things on SignalTools — verdicts on real firms, recovery guidance for real victims — are human-curated. AI is a tool that helps humans work faster on detection patterns and content drafts. Final decisions about what we publish are human decisions.

If you spot an AI-related error on the site (a misclassified message, a false positive verdict, content you believe was generated without sufficient review), email corrections@signaltools.org with the URL and what you noticed. Same response timeline as our regular corrections process (48-hour acknowledgement, 5-working-day resolution).

Last updated: 12 May 2026. This statement is reviewed at least every 6 months or whenever a material change in AI use occurs.