What to Watch Out for in AI-Assisted Shopping Tools

There's a new AI assistant for nearly every kind of shopping decision now — gift finders, meal-plan-to-grocery-list tools, product comparison bots, “what should I buy” chat widgets bolted onto retailer sites. Most of them are genuinely useful some of the time. But testing a handful of them turns up the same three failure patterns often enough that they're worth knowing before you hand one a real decision.

1. Affiliate steering dressed up as personalization

A lot of these tools are funded by commission — they get paid when you buy something through their link, and the easiest program to join usually isn't the best match for you, it's just the easiest program to join. That creates a quiet incentive to nudge recommendations toward whoever pays the referral fee, even when the tool's marketing talks entirely about “personalized to you.” The tell is usually consistency: if every recommendation across very different inputs keeps routing back to the same one or two retailers, that's the commission structure showing through the personalization framing, not a coincidence.

2. Pricing you can't actually calculate in advance

Some AI tools price themselves in “credits” rather than a plain per-use cost. Lindy is a good example — the sticker price for a credit pack looks simple, but converting that into “how much will it actually cost me to do the thing I want” requires guessing how many credits your specific task will consume, which isn't disclosed up front. That's an extra layer of math most software doesn't ask you to do before you've even tried it, and it makes the real price effectively opaque until after you've committed.

3. Confidently describing work it didn't actually do

The third pattern is a gap between what the AI says it did and what it actually produced. Nori, a grocery-planning assistant, is a useful example: it can tell you, in a confident, finished-sounding sentence, that it's built your shopping list — but what actually lands in the list doesn't always match that description. This is arguably the most dangerous of the three failure modes, because a wrong price is usually obvious once you look, while a plausible-sounding summary of work that wasn't actually done properly can slip past unnoticed.

What to check before you trust one

• Can you work out the real cost of a typical use from the pricing page alone — no guessing at credit consumption? If not, that's worth treating as a flag, not a detail to sort out later.
• Do recommendations actually change when your input changes meaningfully, or does everything funnel toward the same one or two options regardless?
• Does the tool show its reasoning (“here's why this fits”) tied to what you specifically told it, or just a generic-sounding result that could've applied to anyone?
• After the tool says it did something — built a list, found a price, compared options — actually check the output against what it claimed, at least the first few times you use it.
• If a number is shown as fact (a price, a savings estimate), is it sourced or clearly labeled as an estimate? A tool that's honest about uncertainty is a better sign than one that states everything with false confidence.