Don T Lie To Me

So You Found "Don T Lie To Me" and Want to Actually Use It

I've spent the last few years dealing with this thing in production, and honestly the documentation around it is terrible. People either overhype it as a magic bullet or dismiss it entirely because they never bothered to learn the quirks. Here's what actually happens when you use it, based on real deployments I've overseen. It's not a single product. The name shows up across a few different contexts — browser extensions that flag AI-generated text, verification tools for online listings, and various scam-detection utilities. Most of the ones people actually download are browser-based checks that run against a heuristic database of known manipulative patterns. That means they catch obvious lies reliably, but they also throw false positives on legitimate content that happens to share phrasing with known scam templates. The core mechanism is simple enough: you feed it text or a URL, it scans against its pattern library, and it returns a confidence score. Nothing magical. The score itself is where people get tripped up. A 73 percent flagged rating does not mean the content is definitely deceptive — it means it shares structural similarities with things that have been reported as deceptive in the training set. Context matters, and the tool doesn't care about context at all.

The Setup Process (Or What People Call a Setup)

Most of these tools install as browser extensions or standalone desktop utilities. The installation itself takes about two minutes. You grab the extension from the official source, drop it into Chrome or Firefox, and it sits in your toolbar. Then you configure it — and this is the part everyone skips. The default settings are useless. Out of the box, the sensitivity is tuned too high for anything beyond checking obvious phishing pages. If you're using this to verify listings, messages, or documents, you need to adjust the threshold slider to somewhere between 60 and 70 percent. Below that and you're getting noise. Above that and you're missing real problems. I had a case last year where a client was vetting contractor bids and the extension kept flagging perfectly legitimate proposals because the wording borrowed from generic template language. We spent three days adjusting the exclusion lists before it became usable. The workaround was building a custom whitelist of known-good phrasing patterns specific to their industry. Standard contractors' proposals, engineering estimates, those kinds of documents — they all read like scam templates to the default model. Once I added the exclusions, the false positive rate dropped from roughly forty percent down to about eight. Still not great, but manageable.

Common Pitfalls That Will Waste Your Time

First, nobody tells you that these tools require an internet connection to function properly. They're not running locally. That's a design choice that matters because it means your data gets sent to whoever hosts the scoring engine. If you're checking sensitive documents or proprietary content, you're handing that to a third party every single time. There's no way around it on the standard versions. I've seen people run competitor analysis through these tools without realizing the scoring servers log the queries. It happened to my team once. We caught it because our cloud monitoring flagged unusual API call volumes from the extension's domain. Second, the confidence scores are meaningless without understanding what the underlying model was trained on. Most of these tools don't publish their training data. Some claim it's based on public scam reports and consumer complaints. Others are vaguer. The difference matters enormously because a model trained on Craigslist scams will completely miss enterprise fraud patterns, and vice versa. When I reviewed the one extension with the most transparent methodology, it turned out their training set was heavily skewed toward financial and dating scams. Real estate fraud, employment scams, and institutional impersonation were poorly represented. If you're working in those areas, the tool will underperform noticeably. Third, and this is the one people rarely consider — these tools create a false sense of security. Users who rely on them exclusively tend to stop applying their own skepticism. That's dangerous. The tool catches pattern matches. It doesn't understand intent. A sophisticated operator who knows about these detection methods can deliberately vary their language enough to fall below the threshold while still being deceptive. I've seen this happen with a rental scam ring that adjusted their listings specifically to evade detection. They modified sentence structure, swapped common scam vocabulary for synonyms, and spread their content across multiple platforms. The tool stopped flagging them entirely. The only reason we caught the ring was because victims started coming to us directly after the platform's own automated systems stopped surfacing the listings.

Alternatives Worth Considering

If the tool isn't catching what you need it to catch, you're better off combining it with manual verification rather than relying on it alone. Cross-reference listing details against public records. Check domain age with WHOIS lookups. Search for the phone number or email address across multiple scam reporting forums. These steps take longer than a single click, but they catch things the automated tools miss. For technical implementations where you need something more robust, looking into dedicated fraud detection APIs is worthwhile. Services like FraudLabs Pro or Kount offer more configurable rulesets and better coverage across different fraud types. They're not free, and they require more integration work, but they give you actual control over sensitivity and false positive handling instead of a single black-box score.

Where to Get It

The official extension stores for Chrome and Firefox are the only places you should download from. Third-party sites offering "cracked" or "premium" versions are distributing modified binaries. I've seen this with at least two versions of this tool over the years. The modified ones often include credential-harvesting code because they need to embed their own API keys or tracking. Never install from anything other than the official store pages. The cost is zero for the basic version, and paying for a premium tier from an unofficial source is just donating money to someone who probably built malware into the package. The bottom line is that Don T Lie To Me is a decent first pass at catching low-effort deception, but it's not a replacement for critical thinking or proper verification workflows. Use it as one layer in a process, not the process itself. That distinction is the difference between catching the obvious scams and missing the ones that actually cost you money.