How Much Would Your Customers Let AI Decide

Part 3 of 4

Asking an AI assistant to compare a few coffee machines is a small ask. Letting it actually choose one, spend your money and sign you up to a service contract is something else entirely. The technology behind both requests might be identical, but what you’re handing over certainly isn’t.

That difference is a good place for for any business thinking about how to bring AI into a customer journey – and it’s really the heart of AI and customer trust: not whether people like AI in general, but what they’re comfortable letting it do in this particular situation, with this particular decision.

Researching something and authorising it are two different decisions

Ipsos’s July 2026 Shopping With AI report found that 27% of AI-aware consumers had used AI to research products, while only 9% reported letting it make a purchase on their behalf. The research covered 7,954 AI-aware consumers across 15 markets in March 2026. It’s worth being precise about what that tells us: these figures describe that particular group of consumers, not shoppers as a whole, and they aren’t a funnel where one number naturally leads to the other.

For a business, the practical point is to look closely at the specific AI buying decision in front of you – the task you’re actually delegating. One customer might be glad of help narrowing down the options, while still wanting to make the final call themselves. Another might be perfectly happy letting a familiar, repeat order go through automatically, provided they’ve set the limits themselves.

There are two relationships worth separating here, too. A customer might consult an independent AI assistant long before they ever reach your website, and there’s only so much you can do about that beyond making sure good information about your business is out there for it to find. But when a customer uses a tool that you provide, you carry direct responsibility for the information it gives, the permissions it operates within, and the support that’s available if something goes wrong.

Look at the commitment behind each task

It helps to think about three different things separately: information, recommendation, and action. These aren’t stages every customer has to pass through in order. They’re simply useful categories to have in mind when you’re deciding what a good conversation with a customer actually looks like.

1. Information is where a tool helps someone understand their options. What’s included? Which model actually meets what they need? Is the information current, and can the customer check it for themselves if they want to?
2. Recommendation is where a tool goes a step further and suggests a choice. What did it take into account to get there? Which trade-offs mattered? And can the customer change one of its assumptions and see a different option come out the other end?
3. Action is where a tool actually does something with real consequences. Can it spend money, change a booking, or agree to terms on the customer’s behalf? What have they actually agreed to, and where does a decision need a fresh, explicit confirmation rather than an assumed one? This is really where customer control in AI lives – not in whether AI is present at all, but in who has to say yes before something happens.

Picture an office manager who’s open to automatic coffee reorders. In this hypothetical, they might be perfectly happy for the tool to suggest a quantity based on what’s been ordered before. They might still want to personally approve a price increase, or a switch to a different product. The question worth asking is simply where that approval sits in the experience you’ve built.

Talking through the real, specific task makes these boundaries far easier to see clearly. Ask your customers what they’d expect to see before agreeing to something, which changes they’d want the chance to review, and what would make them stop and want more information. Their answers will often be different from what the project team assumed going in.

Control includes a clear route to a human

Gartner reported in August 2026 that 50% of the customers they surveyed said generative AI had made their service interactions easier, while 87% still considered access to a human agent essential. Those findings came from 3,566 business and consumer customers surveyed across February and March 2026.

People can genuinely appreciate a fast, automated answer and still want to know someone’s there if that answer doesn’t actually solve their problem. This finding is really about access to a person existing at all, rather than a demand for every single interaction to begin with one.

If a customer’s going to need help, it’s worth explaining that route to them before they’re stuck and frustrated. Be clear about what support can actually do, when it’s available, and what happens to the information they’ve already given you. That’s the real test of trust in automated services – not how quickly the system responds, but how honest it is about what it can actually do. Only ever promise a response time. And only ever promise a response time or a way to fix a mistake that your business can genuinely deliver on.

For the practical work behind building this kind of experience, our guide to preparing an AI sales agent covers approved information and boundaries in detail, and our piece on AI and customer experience looks at the ongoing role human support plays alongside it.

Explain the benefit, and be honest about AI’s role in it

Research by Cicek, Gursoy and Lu, first published online in 2024, found that simply mentioning AI in a product or service description reduced purchase intentions, in experiments involving more than 1,000 US adults. Emotional trust seemed to explain much of the effect, and it was stronger for higher-risk products. Worth noting that the experiments measured intention rather than completed sales.

That points to a genuinely useful question to ask of your own marketing: have you actually explained why a feature helps the customer, or have you just labelled it as “AI-powered” and left them to work out the benefit themselves? It’s worth describing the outcome that actually matters to them, being clear about how AI contributes to it, and being honest about where they remain firmly in control. Hiding AI’s role entirely would just leave a different, equally unhelpful question unanswered.

It’s also worth being careful not to assume customers always prefer a human. Logg, Minson and Moore’s 2019 experiments found that ordinary participants often gave more weight to advice when it was presented as coming from an algorithm than when the same advice came from a person. The specific task and the people involved clearly mattered a great deal, so it’s really a reminder to test this for your own particular decision rather than assume you already know the answer.

Review one decision before you expand the offer

Choose a single customer decision that your AI tool already supports, or might support in future. Write down exactly what it can suggest, what it can actually change, and what genuinely needs the customer’s permission first. Then walk through that experience with real customers and listen closely for any hesitation.

What comes out of that might be a better explanation somewhere in the journey, a clearer confirmation step, or simply an easier way to reach the right person when something isn’t working. It might also reveal that a feature you thought was convenient is actually asking customers to hand over more control than they’re ready to give at that particular point, and that’s just as valuable to know.

If your team needs help choosing practical uses for AI, explore STORY22’s AI training. Start with the customer’s task and the confidence needed to complete it, then decide how the technology should help.

About STORY22 

STORY22 is a full-service international marketing agency based in the UK. As certified StoryBrand guides, we help organisations clarify their message and apply it consistently across marketing strategy, websites, campaigns, content and customer touchpoints. 

Our approach is straightforward: strategy before tactics and clarity before complexity. 

Schedule your free 30-minute strategy call and make sure your marketing is saying exactly what you think it is.