Quick answerYes, in a narrow, clearly defined set of cases: when the customer's stated need falls genuinely outside what your product does well, and staying silent would mean either selling them something that will fail them or leaving them stuck. The policy should not be blanket honesty about every competitor at every turn. It should be a bounded exception, written down, reviewed, and limited to cases where your own product's fit is provably poor for that specific need, not just less optimal than a rival's.
The real question isn't philosophical, it's operational
Founders tend to answer this question in the abstract: "of course we'd never send a customer to a competitor, that's insane" or "of course we'd be honest, trust matters more than one sale." Neither answer survives contact with an actual conversation transcript. The real question is narrower and much more practical: under what conditions, exactly, should the agent say a competitor's feature is the better fit, and who decided those conditions in advance rather than the agent deciding on the fly.
This is distinct from marketing your agent's capabilities honestly, which is about not overselling what your own product does. It's also distinct from differentiating your product once the market has commoditized around the same feature set. This post is about the moment inside a live conversation where a customer describes a need, and the honest technical answer is that a rival's product handles that specific need better than yours does.
Define "genuinely better fit" before the agent has to decide it live
The failure mode isn't an agent that never mentions competitors. It's an agent that decides, in the moment, based on fuzzy pattern matching over a support conversation, whether a rival's tool is "better." That is not a decision an LLM should make ad hoc. It needs a bounded, pre-approved list of scenarios where the answer is already known and written down: specific integration gaps, specific compliance requirements your product doesn't meet, specific scale thresholds where your architecture genuinely underperforms.
Everything outside that pre-approved list defaults to describing your own product's actual capabilities accurately and letting the customer decide, not volunteering a competitor's name unprompted.
From the team
We build production AI systems for startups.
LLM pipelines, RAG, and agent workflows that hold up under real traffic — not just in the demo.
Why this pays off, even though it feels counterintuitive
Sales teams resist this policy instinctively, and the instinct is understandable in isolation. But a customer who signs up for a product that cannot do what they need churns fast, leaves a negative review that costs more in future pipeline than the one deal was worth, and rarely comes back once they've been burned. An agent that says, in a narrow and honest case, "our product doesn't do that well, here is who does" converts that customer into a source of trust rather than a support ticket and a churn statistic. This only works, though, if the exception is rare and specific. If it happens often, the real problem is your product roadmap, not your agent's honesty policy.
What the written policy needs to contain
A workable policy names the specific need categories where competitor mention is allowed, requires the agent to still fully describe what your product can do first, and logs every instance where the exception fires so the product and sales teams can see the pattern instead of discovering it anecdotally. Without that logging, you have no way to notice that the same gap is triggering the exception every week, which is itself a signal someone above the agent's pay grade needs to act on. Track the downstream effect too: a customer who was honestly redirected tends to show up favorably in customer satisfaction and NPS data later, even though they didn't convert that day.
FAQ
Does this mean the agent can freely badmouth or compare against competitors? No. The scope is narrow: acknowledging that a specific need is better served elsewhere, not evaluating or ranking a competitor's overall product.
How often should this actually trigger in a healthy business? Rarely. If it fires often, that is a signal about your product gaps, not a sign the policy is working as intended.
Should the agent ever name a specific competitor by name? Only if legal and sales leadership have pre-approved specific named comparisons. Otherwise, describe the type of solution rather than naming a brand, to avoid becoming an unintentional advertisement or a misrepresentation risk.
Who should own the pre-approved exception list? Product and sales leadership jointly, reviewed on a fixed cadence as your own product's capabilities change.