Selling AI to Skeptical Buyers
Turn this article into takeaways for your work.
Each assistant summarizes the article only for you and suggests best practices for your work.
Selling AI to skeptical buyers means accepting that the doubt is reasonable and answering it with evidence instead of enthusiasm. Most B2B buyers aren't hostile to AI. They've heard big claims, seen some of them fail, and learned that a fluent answer isn't the same as a correct one. A vendor who treats that as an attitude problem to be talked past will lose to one who treats it as a set of specific questions to be answered.
This page breaks "skepticism" into its distinct objection types, shows what kind of proof answers each one, and explains how the sales approach shifts as a result. For demo mechanics and evaluation sets, see why demos matter more for AI products. For messaging, see AI product positioning.
Why Buyers Are Skeptical
Skepticism toward AI is widespread and measurable, and it sits alongside heavy use. KPMG and the University of Melbourne's 2025 global study on trust in AI, covering more than 48,000 people in 47 countries, found that 46 percent are willing to trust AI systems, while 66 percent use AI regularly. People use it and still hold back trust.
The same study found that 70 percent believe regulation is needed. Pew Research's survey of US adults, fielded in August 2024, found 51 percent were more concerned than excited about increased AI use in daily life, and 59 percent had little or no confidence that US companies will develop and use AI responsibly.
Those are general-public numbers, not procurement data, so don't read them as your buyer's exact view. What they show is the starting point: a buyer in a meeting is also a person who has formed these opinions elsewhere.
The Distinct Objection Types
"We're not sure about AI" bundles together several separate worries. Each needs a different answer. The mistake is replying to all of them with the same demo.
Accuracy and hallucination. "What if it's wrong?" The buyer fears confident, fluent errors that nobody catches. KPMG's study gives that fear some grounding: 66 percent of workers surveyed said they rely on AI output without checking its accuracy, and 56 percent said they had made mistakes at work because of AI. Buyers don't need to have read that to sense the risk.
Overclaiming, or "AI washing." "Is this really AI, and does it really do that?" Regulators have started acting on this, covered below.
Data and privacy. "Where does our data go, and does it train your model?" This one often decides whether a deal can proceed at all, because it hits legal and security review.
Failed past pilots. "We tried something like this and it went nowhere." The buyer isn't doubting the technology as much as the vendor's ability to deliver a result in a real workflow.
People and change. "Will our team use it, and does it threaten their jobs?" The economic buyer may be fine while the people who must adopt it are anxious or resistant.
Vendor viability. "Will you still be here in two years?" AI startups are young and the category moves fast. Buyers worry about being stranded on a product that's shut down, acquired or rebuilt.
Regulators Made Overclaiming a Legal Risk
Buyers' suspicion of inflated AI claims isn't only a mood. Two US regulators have taken enforcement action on it.
In March 2024, the SEC announced settled charges against two investment advisers, Delphia (USA) Inc. and Global Predictions Inc., for making false and misleading statements about their use of AI. The SEC fined them $225,000 and $175,000, a combined $400,000. SEC Chair Gary Gensler said investment advisers should not mislead the public by saying they are using an AI model when they are not. This is the origin of "AI washing" as an enforcement term.
In September 2024, the FTC announced Operation AI Comply, a sweep against deceptive AI claims that named DoNotPay, Ascend Ecom, Ecommerce Empire Builders, Rytr and FBA Machine. The release states there is no AI exemption from the laws on the books.
For a founder, the takeaway is practical. Every capability claim in a sales deck, website or email is something a buyer's counsel may read closely, and one you should be able to support with a test result. If you can't show that the product does a thing, don't say it does.
What Evidence Answers Which Objection
Claims don't move a skeptic, but a specific artifact can. The table maps each objection to the proof that tends to address it. The right column describes categories of evidence, not a promise of results.
| Objection | What the buyer is really asking | Evidence that answers it |
|---|---|---|
| Accuracy and hallucination | How often is it wrong, and how would we know? | A measured error rate on a stated test set, a visible failure case, and the review step built into the workflow |
| Overclaiming | Can you prove each capability you described? | A claims list where every item maps to a test, plus plain statements of what the product doesn't do |
| Data and privacy | Where does our data go and who can see it? | Written data-handling terms: retention, use for training, subprocessors, deletion, security documentation |
| Failed past pilots | Why would this one reach production? | A scoped pilot with written success criteria, a named owner and a decision rule; see from pilots to contracts |
| People and change | Will anyone use it, and what does it do to roles? | A workflow where people stay in control, a rollout plan with training, and honest framing of which tasks change |
| Vendor viability | What happens if you disappear or change course? | Customer references, data export rights, contract continuity terms |
Honest Limits Build More Trust Than Confidence
The instinct under skepticism is to sound more certain. That usually backfires, because certainty is exactly what the buyer distrusts.
State the limits first. Say what the product is good at, what it handles poorly, and how you know. "It reaches this accuracy on this kind of input, it struggles with that kind, and those go to a person" gives a buyer something to check. "It's highly accurate" gives them something to doubt.
Describe the controls, not just the failures. A limit with a safeguard (low-confidence outputs routed to review, an audit trail, an easy way to override) is a risk the buyer can manage. A limit with no safeguard is a reason to wait.
Use the vocabulary buyers' own risk teams use. The NIST AI Risk Management Framework lists the characteristics of trustworthy AI as valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. It also says creating trustworthy AI requires balancing these characteristics based on the system's context of use. It's a voluntary framework, not a certification, so don't claim "NIST compliance." But organizing your answers to a security or risk questionnaire around those headings shows you know what a careful buyer is checking.
Human Oversight as a Product Feature
For most skeptical buyers, the question isn't whether AI can replace a step, but whether a person remains accountable for it. Designing and selling the product with oversight visible changes the conversation.
- Show where a person decides. Draft-and-approve flows, confidence thresholds and escalation paths make the human role concrete.
- Keep an audit trail. Buyers in regulated or customer-facing settings need to see what the system did and why, after the fact.
- Let the buyer set the autonomy level. Starting in suggestion mode and moving to automation as trust builds lets a cautious buyer say yes sooner.
- Name who is accountable. "The reviewer signs off" is a clearer answer than "the AI handles it."
This also speaks to the people-and-change objection. A tool framed as support for a team's judgment meets less resistance than one framed as a replacement, and it's usually more accurate to describe it that way too, since most deployments keep people in the loop.
References and Proof From Peers
Skeptics trust peers more than vendors. A reference call with a customer in a similar role, size and workflow is among the strongest assets you can offer, because the buyer gets answers from someone with no stake in the sale.
A few practices make references work harder:
- Match the reference to the objection. A buyer worried about data should talk to a customer who went through security review. A buyer burned by a past pilot should talk to one who went from pilot to production.
- Prepare references for honest answers. If a reference only praises, it reads as scripted. A reference who says "it needed three weeks of tuning, and then it worked" is more credible.
- Don't invent proof. Only cite results you can substantiate, with the method behind them, in line with the enforcement actions above.
Early-stage companies without many references can offer smaller forms of proof: a design partner willing to speak, a public evaluation method, or a pilot structured so the buyer sees the evidence before committing. Getting those buyers into a decision process is covered in the AI buying committee.
Contractual Commitments Close the Gap
Some doubt can't be removed by evidence, because it concerns the future. Contract terms can carry that residual risk. Buyers can't verify tomorrow's performance, but they can ask you to commit to something about it.
Terms that often come up with skeptical buyers include:
- Data handling: no training on customer data without permission, defined retention, deletion on exit
- Performance or service commitments tied to the measured criteria from the pilot
- An exit path: the right to export data and end the contract if agreed criteria aren't met
- Notice of material changes to models or subprocessors
- Security documentation and audit rights proportionate to the data involved
You don't need to concede everything. You need to know which commitments you can honestly keep, offer those, and decline the rest plainly. A commitment you can't meet is worse than none. Structuring the path from a successful test to these terms is covered in AI pilots to contracts, and building the financial case alongside is covered in the AI product ROI business case.
How the Sales Approach Changes
Selling to skeptics isn't a different funnel, but it shifts emphasis at each point.
- Diagnose the objection before answering it. Ask which worry is driving the hesitation. The questions in the table above are a starting list. The classic discipline in objection handling applies: listen, clarify, then respond with proof.
- Lead with evidence, not claims. Open with a test result, a failure case or a reference, not a feature list.
- Say what you don't do. Naming a limit early moves you from "vendor" to "someone who has thought about this."
- Bring the risk owners in early. Legal, security and the people who'll use the product often decide whether a deal survives. Don't wait for them to show up late. See champion-based selling for equipping an internal advocate to carry your case.
- Teach where it helps. A buyer who doesn't yet know what good looks like may benefit from a clear point of view on how to evaluate AI. The Challenger Sale covers taking that stance without being pushy.
A Quick Self-Check Before Your Next Call
- Can we support every capability claim we make with a test or a customer?
- Do we know our measured error rate, and how we measured it?
- Can we explain, in plain terms, where the buyer's data goes?
- Where does a person review or override the output?
- Which references fit this buyer's specific concern?
- What contractual commitments can we actually keep?
Key Facts
- Trust trails use: a 2025 KPMG and University of Melbourne study of 48,000+ people in 47 countries found 46 percent willing to trust AI systems, 66 percent using AI regularly and 70 percent saying regulation is needed.
- Unchecked output is common: the same study found 66 percent of workers rely on AI output without checking it, and 56 percent reported making mistakes at work due to AI.
- US public caution: a Pew Research survey fielded in August 2024 found 51 percent more concerned than excited about AI in daily life and 59 percent with little or no confidence in US companies to use AI responsibly.
- AI washing is enforced: in March 2024 the SEC fined Delphia ($225,000) and Global Predictions ($175,000) for false statements about AI use.
- No AI exemption: the FTC's September 2024 Operation AI Comply named five companies and stated there is no AI exemption from existing law.
- A shared vocabulary exists: the NIST AI RMF 1.0 lists seven characteristics of trustworthy AI, from valid and reliable to fair with harmful bias managed.
Related Reading
Co-Founder & COO
On this page
- Why Buyers Are Skeptical
- The Distinct Objection Types
- Regulators Made Overclaiming a Legal Risk
- What Evidence Answers Which Objection
- Honest Limits Build More Trust Than Confidence
- Human Oversight as a Product Feature
- References and Proof From Peers
- Contractual Commitments Close the Gap
- How the Sales Approach Changes
- A Quick Self-Check Before Your Next Call
- Key Facts
- Related Reading