How to Spot AI Washing Before You Add Another Tool to Your Stack

How to Spot AI Washing Before You Add Another Tool to Your Stack

Every tool in the directory says it's powered by AI now. Half of them are telling the truth, and half of them ran a find-and-replace on their old marketing copy. AI washing is what happens when a company slaps an AI label on a feature that was never intelligent to begin with.

It has gotten good enough to fool people who should know better. Spotting the difference early saves a subscription no one ends up using.

What "AI Washing" Means and Why It's Suddenly Everywhere

The term comes from the same place as greenwashing. A company wants credit for something it has not built, so it borrows the language of a trend that already has trust. AI washing works the same way.

A scheduling feature gets called an AI assistant. A basic filter gets called machine learning, and a template picker gets called generative design.

Using inflated language is not automatically illegal. Marketing has always stretched features to make them sound bigger than they are. The stakes change when an AI claim becomes false or misleading and buyers are making decisions based on it.

What a Real AI-Washing Case Looks Like

This has already moved beyond a hypothetical marketing problem.

In March 2024, the U.S. Securities and Exchange Commission charged investment advisers Delphia and Global Predictions with making false and misleading statements about how they used artificial intelligence. The firms agreed to settle the charges and pay a combined $400,000 in civil penalties.

According to the SEC, Delphia had claimed it used AI and machine learning with client data to inform its investment process. The SEC found that the company did not have the AI and machine-learning capabilities it had represented. Global Predictions was also cited for claims including calling itself the "first regulated AI financial advisor" and promoting "AI-driven" forecasts.

That is the problem in its clearest form. The issue was not that the companies used ordinary automation. It was that specific AI capabilities became part of the pitch, while the underlying technology did not match what customers were being told.

The Four Claims Every "AI-Powered" Tool Makes

Strip away the branding, and most AI-powered tools are making one of four claims: reading your intent, learning from your behavior, generating something new, or predicting what happens next. Only some of those claims require a trained model to be true.

Reading intent and predicting outcomes are two areas where model-based systems are common. "Learning from behavior" is murkier because the phrase can describe actual machine learning or simple rules presented as personalization.

Generating something new is also easy to exaggerate. A template pulled from a library can look like generated output at a glance, especially when the buyer only sees a polished demo.

"Powered by AI" vs. Powered by an If-Statement

A rule that says "if the post gets fewer than ten likes in an hour, suggest a new time" is a threshold someone wrote once, not a trained system making a judgment call.

That kind of automation is often the right tool for a small, well-defined problem. The trouble starts when the vendor markets it as a system that adapts on its own.

When "Machine Learning" Just Means a Spreadsheet Formula

Some tools call a weighted average machine learning because the math technically involves numbers being combined.

Machine-learning systems are trained on data and can be updated or retrained as new data becomes available. A fixed formula does the same calculation tomorrow that it did today unless somebody changes it.

A formula that stays fixed after launch is not learning anything, regardless of how many times the phrase "machine learning" appears on the pricing page.

A Five-Minute Test to Spot AI Washing Before You Buy

Most vendors will not lie outright if asked a direct question. They will often avoid answering it if the honest answer weakens the pitch.

A five-minute conversation, or a careful read of the documentation, usually reveals which side of the line a tool sits on.

Ask for the Specific Model, Not the Buzzword

Ask what model or technique sits behind the feature, whether it's a specific type of neural network, a named large language model, or a decision tree.

A team that built real AI capability can usually answer in one sentence. A team that added a thin AI layer on top of existing software is more likely to answer with a mission statement.

Give It an Edge Case and Watch What Happens

Feed the tool something unusual: a post in a second language, a niche industry term, or a request outside its main use case.

A capable AI system should still produce something relevant when the input moves away from its ideal demo, even if the quality drops. A rigid rules system carrying an AI label is more likely to hit a boundary the developer never anticipated and return something generic or nothing at all.

The failure itself matters less than what it reveals about the mechanism underneath.

Real Questions to Ask a Vendor, Not Just Their Website

Companies write marketing pages to close a sale, not to answer a technical question honestly. A sales call or support chat gets closer to the truth.

Find out how the AI feature performed for a customer whose use case differs from the demo. Push on what happens when the underlying model gets something wrong, and whether a person reviews it.

Then ask how the tool improves over time. Does that come from retraining or updating a model with new data, or from somebody manually changing a rules file when a new situation appears?

What Genuine AI Capability Looks Like

A comparison worth trusting names the specific capability being tested rather than repeating a vendor's own description of itself.

One breakdown of social media AI tools separates tools using capabilities such as sentiment analysis and generative drafting from products that mainly added a chatbot layer to a basic scheduler. It says plainly which is which for each entry.

Building a Habit So AI Washing Stops Working

The fastest way to stop falling for AI washing is to ask the same three questions every time.

What specific model or technique powers this? What happens when it is wrong? How does it get better over time?

If a vendor answers all three without retreating to marketing language, the AI claim is probably real. If an answer turns vague, the label may be doing more work than the feature.

Questions About Spotting AI Washing

What is AI washing?

AI washing is when a company markets a feature as AI-powered when it's really just basic automation or rules written by a person. The term follows the same pattern as greenwashing, where a label gets used for credibility it has not earned.

How can I tell if an AI feature is real?

Ask what specific model or technique sits behind the feature, and see if the answer names something concrete.

Then test what the system actually does. Repeated prompts, edge cases, and questions about how the system improves can reveal whether the feature behaves like a model or follows a much narrower set of predefined rules.

Does AI washing mean a tool is bad?

No. A tool built on plain automation can still solve a real problem well.

The issue is the mismatch between what a tool does and what its marketing claims it does. A basic scheduler that is honest about being a scheduler is a better buy than one pretending to be an AI strategist.

The Best Filter Is Still a Direct Question

AI washing works because most buyers do not ask what sits behind the label. The fix is asking the same three direct questions every time a tool claims AI capability, without needing to become a machine learning expert.

Pay attention to which answers turn vague. A directory with ten thousand tools will always have some that stretch the truth. The five minutes it takes to check is cheaper than a subscription that never does what it promised.