There are a lot of factors to consider when it comes to choosing a car and how it will work for your everyday life. A buyer might need space for a growing family while wondering whether an electric vehicle would work without a charger at home. An AI assistant can help organise that research, explain the options and link to reviews and owner discussions supporting its recommendations.
Those links give buyers a way to check the advice. But has the assistant understood what it read? Do the sources actually support its answer? Companies building or deploying shopping assistants need to check both before asking customers to trust the recommendation.
For our UK Automotive AI Visibility Index, we analysed 7,334 AI answers containing 56,411 citations. We collected the answers on 13 October 2026, using unbranded UK car-buying prompts across ChatGPT, Google AI Mode, Google AI Overviews and Microsoft Copilot.
Our companion social-citation study showed how differently the services cite social content. Social platforms supplied 21.9% of Google AI Overviews’ citations, compared with 1.0% of ChatGPT’s. Across the whole sample, social accounted for 4.8%.
The sources behind an answer offer different perspectives on a car. An owner discussion can describe long-term use, while a road test assesses the vehicle under particular conditions. Which is useful depends on the buyer’s question. For companies assessing shopping assistants, the task is to check whether the assistant uses those sources appropriately for the buyer’s question.
Does each claim have support?
Finding a credible source is only part of producing a reliable answer. A link to a respected motoring publication tells you where to look. You still need to read the review and check what the assistant has taken from it. Does it back up the statement, or simply discuss the same car? Has the assistant left out a qualification that changes the meaning?
It’s worth checking each claim that could affect the buying decision. The assistant might get the specifications right, then recommend the car for a reason its sources never address. That reason needs checking just as closely as the figures.
The format of a citation matters, too. About 52% of cited YouTube links pointed to a specific timestamp. Watch that section and enough of the surrounding video to understand what the reviewer means. A relevant video title alone won’t tell you whether the claim holds up.
Check the model year, trim and powertrain, too. For UK recommendations, the evidence needs to concern the UK version. Otherwise, the assistant could accurately describe a different car from the one it is recommending.
Does the answer explain its limits?
A manufacturer’s specification sheet and an owner’s account can both be useful, but they tell us different things. The answer should make clear whose experience or assessment it is drawing on. One owner reporting a problem doesn’t tell us how common that problem is.
Two reviews can disagree without either being wrong. The reviewers may have tested different versions or placed more weight on different aspects of the car. When that happens, does the assistant explain what they disagree about? Or does it present a firm conclusion with links that tell different stories?
Then there’s the buyer. Why does the assistant think this car suits them? Teams need to check whether that explanation reflects what the person actually asked for, including any assumptions the assistant has made.
Sometimes the assistant needs to ask another question before recommending anything. Sometimes it needs to explain that the evidence is mixed. A useful answer should make those gaps clear, so the buyer knows what still needs checking.
Test the advice before customers rely on it
Companies should be able to answer all these questions before putting a shopping assistant in front of customers. I recommend starting with a consistent set of car-buying prompts covering different requirements. For each answer, pick out the claims that could change someone’s decision. Find the passage supporting each one and record it alongside the claim. Check whether it describes the right vehicle and includes any conditions the assistant has left out. Verify current UK specifications and prices, recording when you checked them.
Recording the problem gives the team something specific to correct and test again. Include prompts that leave important buyer requirements unstated, so you can assess whether the assistant asks for clarification. Use the same checks when comparing services. Assess how well each answer supports its recommendation as the number of references alone won’t tell you that.
Repeat the checks when models or search features change substantially. Keep the answers that exposed earlier problems and test those cases again. Finding that an error has disappeared once isn’t enough to establish that it stays fixed.
AI can help assemble research and explain alternatives. To trust its recommendation, a buyer needs an explanation backed by evidence about the car they can actually purchase. Companies deploying shopping assistants should check that connection before leaving customers to discover where it breaks.
