Every vendor says they use AI. The phrase means nothing. In 2026, “we use AI” is a signal of nothing, and treating it as a signal is how buyers get had.
Open any vendor’s product page. Scroll past the navigation, past the hero, past the testimonial. Somewhere in the second or third section, the word AI appears. Sometimes it is in a headline. Sometimes it is in a feature list. Sometimes it is in a footer link to the AI whitepaper. The product, in every case, “uses AI.” The phrase has become the marketing equivalent of the words “cloud native” circa 2015, or “mobile first” circa 2010. The phrase is everywhere. The phrase means everything and nothing.
Some of the vendors that use the phrase are using AI in a real sense. The product’s behaviour is materially different because of the AI. The feature would not exist without the model. The user experience is shaped by the model’s outputs. Some of the vendors are using AI in a partial sense. The product has AI features, but the core product is unchanged. Some of the vendors are using AI in a marketing sense. The product is a rules engine with the word AI in the documentation. Some of the vendors are using AI in a negative sense. The product is worse because the AI is bolted on, and the user is paying for the AI rather than for the underlying product.
All four groups say they use AI. The phrase, by itself, does not distinguish them.
The four flavours of “we use AI”
Flavour one is genuine AI. The product’s core behaviour is driven by a model. The product would not work, or would not work well, without the model. Examples, in 2026, include GitHub Copilot for code generation, Notion AI for document assistance, Grammarly for writing assistance, Harvey for legal research, Glean for enterprise search. The product is, in some sense, the model, with a UI on top. The phrase “we use AI” is accurate. The phrase understates the case.
Flavour two is AI augmented. The product has a rules based core, with AI features bolted on or integrated. The product worked before the AI. The product works better with the AI. The product would still work without the AI, just less well. Examples, in 2026, include most CRM systems (AI assisted lead scoring on top of a relational database), most marketing automation tools (AI assisted copy generation on top of a campaign engine), most customer support platforms (AI assisted answer suggestion on top of a ticket system). The phrase “we use AI” is accurate. The phrase overstates the case.
Flavour three is AI in marketing only. The product is a rules engine, a database, a workflow tool, with no AI in the runtime behaviour. The product’s documentation, marketing, or roadmap mentions AI. The mention is, in many cases, about a future feature, a beta, a “coming soon.” The product is sold as AI enabled. The product is, in practice, not. The phrase “we use AI” is misleading. The product is not AI in any meaningful sense.
Flavour four is AI in name only, and the product is worse for it. The product has been retrofitted with AI features that the underlying product does not need. The features are slow. The features are expensive. The features are often wrong. The product would be better without the AI, but the marketing required the AI, so the AI was added. The phrase “we use AI” is accurate. The phrase should be a warning.
How to tell them apart
The honest test is to ask the vendor to describe, in concrete terms, what the AI does, what data it uses, what the user controls, and what the failure modes are. The test counts as the same one a good security review would use, applied to a feature claim. The vendor’s answer to the four questions, in detail, tells you which flavour you are buying.
If the answer is “the AI reads your customer interactions and predicts the next best action, with a confidence score, retrained weekly on your data, with a fallback to the previous rule based system when the confidence is below 0.6”, you are looking at flavour two. The product has a real AI feature, integrated thoughtfully, with the right safety properties. The phrase “we use AI” is accurate. The product is a normal product with an AI feature.
If the answer is “the AI is built into the product. It just works. You don’t need to think about it.”, you are looking at flavour three or four. The vendor cannot describe the AI because the AI is not a specific thing, the AI is a marketing position. The product is either rules based with AI marketing, or rules based with bolted on AI features. The phrase “we use AI” is misleading.
If the answer is “the model takes your prompt and generates a response, with the model’s own safety controls and no custom tuning”, you are looking at flavour one, with the model wrapped in a thin UI. The product sits as the model, the product is not adding much value over the model itself, and the price premium is for the UI rather than the AI.
The questions to ask
- What does the AI, specifically, do in this product? What counts as the input, what runs as the output, what amounts to the model doing between them?
- What data does the AI see? Customer data, anonymous data, public data? What is sent to the model provider? What is stored? What is logged?
- What does the user control? Can the user turn the AI off? Can the user override the AI’s output? Can the user see the AI’s reasoning?
- What are the failure modes? What happens when the AI returns nothing? What happens when the AI is wrong? Is there a fallback to a non AI path?
- What sits as the latency cost? Does the AI add seconds, minutes, or hours to the workflow? Is the user waiting on the AI?
- What sits as the dollar cost? Is the AI billed per call, per user, per month, or included? Is the cost predictable or does it scale with usage?
- Can the AI be replaced? If the vendor switched off the AI feature tomorrow, would the product still work, just less well? Or would the product stop working entirely?
These seven questions, asked of a vendor in a sales call or a security review, will distinguish flavour one from flavour two from flavour three from flavour four in 10 to 15 minutes. The vendor’s ability to answer them, in detail, sits as the signal. The vendor’s inability to answer them, or the vendor’s answer that consists of marketing phrases, is also a signal.
The proof to look for
Beyond the questions, the proof is in the product. The proof of flavour one is a feature that would not work without the model. The proof of flavour two is a feature that works better with the model, and a clear explanation of what the model is doing. The proof of flavour three runs as the absence of a model in the runtime behaviour, no matter what the marketing says. The proof of flavour four is a feature that is slow, expensive, often wrong, and not integrated with the rest of the product.
The test you can run, as a buyer, is to use the product for a week. Use the AI features. Use the non AI features. Compare the experience. If the AI features are noticeably better, the product is flavour one or two. If the AI features are noticeably worse, the product is flavour four. If the AI features are indistinguishable from the non AI features, the product is flavour three, and the AI is marketing.
The case for and against AI in a product
The case for AI in a product is real. Some features are genuinely better with a model. Code completion, document assistance, search, summarisation, content generation, all of these are places where the model is doing work that the rules based system could not do. The companies that have integrated the model well have built products that are meaningfully better than the rules based versions. The work is real, the value is real, the AI is real.
The case against AI in a product is also real. Some features are worse with a model. Features that require deterministic behaviour, that require audit trails, that require predictable latency, that require cost predictability, all of these are places where the model is a poor fit. The companies that have bolted the model onto features that did not need it have built products that are slower, more expensive, less reliable, and less useful than the rules based versions. The work is real, the value is negative, the AI is a tax.
The judgement, for a buyer, is to know which is which. The judgement is, in 2026, the most important vendor evaluation skill there is. The phrase “we use AI” sits as the start of the conversation, not the end.
The bottom line
“We use AI” is a phrase that means everything and nothing. The phrase tells you that the vendor has a marketing page. The phrase does not tell you whether the AI is doing the work, augmenting the work, or being used to inflate the price. The buyers who treat the phrase as a signal are being had. The buyers who can tell the difference are buying the right product.
The difference sits as the seven questions. Ask them. Listen to the answers. Demand detail. The vendor with a real AI feature will answer in detail. The vendor with a marketing position will not. The answer sits as the signal. The phrase serves as the noise.
Sources & Further Reading
All claims in this article are sourced from primary documentation, vendor advisories, and reputable security researchers.
Spotted an error? Email the editor. Corrections are issued with a visible correction note.
Editorial standards. Every article on humanrequired.org is reviewed by a human editor before publication. AI may assist with drafting or research; final editorial control is human. Read the full standards.
