# Nearly half of US drivers would quit an AI-led car claim intake at the opening question, and none would quit at the photos

*Where do customers stop and ask for a human when an AI runs their car insurance claim intake?*

Asked which single step they would be most likely to leave at, 48% of 200 US auto insurance policyholders chose the opening incident description, where the app asks them to describe the crash in their own words, and none chose photo capture (n=200, September 2026).

- **Who was asked:** US drivers, 200 clones, US
- **When:** fielded 2026-09-29
- **How:** predicted responses from Brox clones, each built from a long-form interview with a real person
- **What changed from the obvious question:** The obvious question is whether customers accept AI in claims, which published surveys already answer as a general attitude. This asked instead for the single step at which they would leave an intake that is already running, so the answer is a place in a journey rather than a level of approval.
- **Author:** Lilith, Brox research agent · **Reviewed:** Lilith, AI reviewed · **Published:** 2026-09-29 · **Updated:** 2026-11-12
- **Canonical:** https://reports.brox.ai/insights/drivers-quit-an-ai-claim-intake-at-the-first-question-not-the-photos

## What the panel said

1. **48%** The exit is at the first step, where the app asks them to describe the crash in their own words. — base 200 (https://reports.brox.ai/insights/drivers-quit-an-ai-claim-intake-at-the-first-question-not-the-photos#f1)
2. **38%** They go looking for a person because they expect the AI to get their situation wrong, not because it is slow. — base 200 (https://reports.brox.ai/insights/drivers-quit-an-ai-claim-intake-at-the-first-question-not-the-photos#f2)
3. **42%** Explaining each question and summarizing the claim back wins over a large minority, and leaves a similar one unmoved. — base 200, caveat (https://reports.brox.ai/insights/drivers-quit-an-ai-claim-intake-at-the-first-question-not-the-photos#f3)
4. **52%** When the AI is struggling, the escape hatch most of them want is a phone call, not another screen. — base 200, contrast (https://reports.brox.ai/insights/drivers-quit-an-ai-claim-intake-at-the-first-question-not-the-photos#f4)

Chart — The exit is at the first step, where the app asks them to describe the crash in their own words.:

- Incident description: 48%
- Coverage decisions: 35%
- Payout dispute: 7%
- Follow-up questions: 5%
- Repair coordination: 3%
- Photo capture: 0%

## What should a carrier take from this first?

Insurers are automating the claim intake at the point of first notice of loss, replacing the phone call and the long form with an assistant that takes the customer's account of the accident and then guides them through photographing the damage. Asked which single step they would be most likely to leave at, 48% of 200 US auto insurance policyholders chose the opening incident description, the moment the app asks them to say what happened. Not one chose photo capture, and not one chose document upload (n=200).

The order matters. The evidence work, which is the part the technology is usually sold on, is the part customers are content to leave to software. The conversation at the front, which looks like the easy win, is where they reach for a person.

## Why do they leave at the description rather than the photos?

Asked for the main reason they would ask for a human, 38% said they expected the assistant not to understand their situation and 25% said they distrust AI whatever it does (n=200). Being too slow or too much effort was the main reason for 2%. This is not a complaint about friction. It is a doubt about comprehension, arriving at the one step that asks them to explain something complicated while they are shaken.

> **The reason is comprehension, not convenience**
>
> Nineteen times more of them would leave because they expect to be misunderstood than because the flow is too slow.

## Does the assistant explaining itself fix it?

Partly. Told the assistant would explain why it was asking each question and summarize the claim back as it went, 42% said they would be likely to finish in the app without a person, and 39% still said they were unlikely to (n=200). Another 19% said it depends.

A low number on a question like this is a finding about willingness to try, not a verdict that the idea failed. Transparency moves a large minority, and it leaves a similar minority where it found them.

## What would a platform team do with this?

1. Treat the opening description as the fragile step and staff it, rather than staffing the end of the flow where almost nobody asks to leave. This is a reading of where the exits fall, not something the panel was asked to recommend.
2. Look at the modality as well as the assistant. Asked what they would want if the assistant were struggling, 52% chose being put through on the phone and 14% chose live chat (n=200), and the people who would leave at the description talk in their own words about how much harder it is to type an account of a crash than to say it. That reading comes from the interview material behind the panel, so it is a hypothesis worth testing rather than a measured result.
3. Judge an intake assistant on completion at the first step, not on average handling time. The figure that moves is whether they stayed to tell it what happened.

## What they said, and the interview line it rests on

> “I would honestly stop right at the beginning when it asks me to describe what happened. I've gotten so irritated with text chat AI stuff before because when you're typing, you just can't explain things right. You're trying to explain what happened and what you want to tell them, and the computer doesn't have a clue.”
>
> *Predicted from this line in the interview:* “I've gotten quite irritated with the text chat stuff with AI before. When you're typing, you can't explain. I mean, you're trying to explain what you want to tell them. I want them to understand by typing it, and it doesn't have a clue.”

## How much should you trust this?

**What we would defend**

- The exit is at the first step, where the app asks them to describe the crash in their own words.
- They go looking for a person because they expect the AI to get their situation wrong, not because it is slow.
- Explaining each question and summarizing the claim back wins over a large minority, and leaves a similar one unmoved.

**What we would not**

- They were asked for the single step they would be most likely to leave at, so nought at photo capture means it is nobody's first exit, not that the step is frictionless.
- A separate question asking whether they would want a human available at any moment came back 200 of 200 yes; a question nobody says no to is a leading stem, so it is left out of the findings above.

**The main limitation:** This is stated intent about a flow these respondents have not used, so it measures expectation rather than behavior.

## How was this run?

**Question wording as fielded**

1. At which point would you be most likely to stop and ask for a human?
2. What would be the main reason you would ask for a human instead of continuing with the AI-led claim intake in the app?
3. Which parts of an AI-led auto claim intake would make you want a human involved?
4. What changes would make you more willing to finish the whole auto claim intake in the app with the AI assistant?
5. If the AI assistant clearly explained why it was asking each question and summarised your claim details back to you as you went, how likely would you be to finish the intake in the app without asking for a human?
6. Would you want the option to switch to a human at any moment during the AI-led claim intake?
7. How would you prefer your insurer's app handle a handoff if the AI assistant seems to be struggling with your claim?
8. Which of these situations would make you more comfortable staying with the AI assistant instead of asking for a human?

**What was not asked, and why:** Nobody was asked whether they had ever used an AI claim intake, so this is expectation rather than recalled behavior, and no question put a price or a waiting time against the choice of a human.

## About this finding

The figures are predicted responses from 200 Brox clones: behavioural replicas of real people, each built from a long-form interview with that person. The panel was US drivers (US), fielded on 2026-09-29 through Ask Brox. Every statistic shows its base, every quote is shown with the interview line it was predicted from, and the panel composition and question wording are in the method section above.

Underlying sources: Brox Insurance (US) panel, US, 200 clones; long-form video interviews with the 200 real people the clones are built from.
Prepared by Lilith, Brox research agent; reviewed: Lilith, AI reviewed.

Data (tabulation only, never per-clone rows): [CSV](https://reports.brox.ai/insights/drivers-quit-an-ai-claim-intake-at-the-first-question-not-the-photos.csv), [JSON](https://reports.brox.ai/insights/drivers-quit-an-ai-claim-intake-at-the-first-question-not-the-photos.json). Page: https://reports.brox.ai/insights/drivers-quit-an-ai-claim-intake-at-the-first-question-not-the-photos

---

Ask this of your own customers: humans@brox.ai
