---
title: "How to measure the return on AI search work · Avouch"
description: "Measure the return on AI search work using recorded visits, customer feedback and sales, with clear costs and an honest account of the gaps."
url: https://avouchedfor.com/insights/measure-ai-search-roi
markdown_url: https://avouchedfor.com/insights/measure-ai-search-roi.md
---

# How to measure the return on AI search work

See what your AI search spending brings back, and where the numbers leave gaps.

**Steen Stones** · Reviewed 8 Oct 2026

Compare your AI search costs with the business you can trace back to it. Put recorded visits beside what buyers tell you, since each misses part of the journey.

## What should go into the cost?

Count the work you pay for, including content, technical work, tools and team time. Ahrefs separates spending specific to AI search from costs shared with wider search and content work.

Show those shared costs in your report, so your comparison includes the work that helped produce the results.

Keep shared costs visible

Swipe across to read all columns.

| Source | Cost | What to record |
| ---------------------------------------------------------------------------- | -------------------------- | -------------------------------------------------------- |
| [Ahrefs: measuring AI search return](https://ahrefs.com/blog/ai-search-roi/) | Content and technical work | The work paid for, including costs shared with wider SEO |
| [Ahrefs: measuring AI search return](https://ahrefs.com/blog/ai-search-roi/) | Tools and team time | The resources used for the work |

## What can your website reports tell you?

Start with the sales you can link to visits from AI tools, which Ahrefs calls tracked revenue. Some buyers return later through a search for your name or by typing your address, leaving a gap.

Google calls visits with no clear source direct traffic, but that does not mean an AI tool sent them. Typing your address, following a document link or using an ad blocker can affect how a visit is recorded.

Keep unexplained visits separate. Calling them all AI visits would claim more than Google's definition tells you.

## What do buyers say brought them to you?

Ask new customers how they found you, then compare their answers with the visits your website recorded.

Graphite reports that n8n compared its last recorded website source with a survey after signup. Analytics credited roughly 1% of conversions to AI answer engines, while the survey credited roughly 9%.

Those are n8n's results, so use the comparison rather than applying its gap to your sales. Report your survey and website figures separately, since adding them risks counting the same person twice.

Different methods gave n8n different answers Share of conversions attributed to AI answer engines. These are n8n’s results, not a forecast for your company.

- Last recorded website source 1%
- Customer survey 9%

### Source and study details

Graphite reports roughly 1% in GA4 and roughly 9% in n8n’s survey. Read 8 October 2026. The methods can overlap; do not add the figures.

The chart’s values, kept separate

Swipe across to read all columns.

| Source | Measurement | Approximate share |
| --------------------------------------------------------------------------------------------------- | ------------------------ | ----------------- |
| [Graphite: n8n attribution research](https://graphite.io/five-percent/research/n8n-attribution-gap) | GA4 last recorded source | 1% |
| [Graphite: n8n attribution research](https://graphite.io/five-percent/research/n8n-attribution-gap) | Survey after signup | 9% |

## How do you connect this to business results?

Kevin Indig recommends comparing exposure, behaviour and business outcomes. Put where you appear beside what buyers do and the business that follows.

When those measures rise together, the evidence grows stronger. As Indig points out, that still does not prove your AI search work caused the rise.

George Bonaci makes a useful point: keep a way to assign credit, with the company's goals in view.

> You will still need an attribution model and you will still need org level goals.

George Bonaci, quoted in Growth Memo The usefulness of attribution for AI Search

[Read original from George Bonaci, quoted in Growth Memo (opens in a new tab)](https://www.growth-memo.com/p/the-collapse-of-attribution)

## What should the next report answer?

Show what you spent, the visits you traced, what customers told you and the business results. Label each measure clearly so the reader can see its limits.

Assigning credit to a source and showing that the work changed the outcome answer different questions. Indig calls the latter incrementality, and assigning credit alone does not establish it.

## Read next

How to check your company’s AI search visibility

[Read the guide ](https://avouchedfor.com/insights/check-company-ai-search-visibility)

## Common questions

### Does direct traffic mean someone found us through AI?

No. Google uses direct traffic when it has no clear referral source. It can include people typing your address or following links in offline documents.

### Should we stop using website reports?

Keep website reports, then compare the visits and sales they record with customer feedback and wider business results.

### Can more AI visibility prove a return?

Looking at visibility and business outcomes together strengthens the evidence. It does not, on its own, prove that one caused the other.

## Sources

1. [ahrefs.com, AI Search ROI Is Messy. Here's How to Measure It Anyway (opens in a new tab)](https://ahrefs.com/blog/ai-search-roi/), checked 8 October 2026
2. [growth-memo.com, The usefulness of attribution for AI Search (opens in a new tab)](https://www.growth-memo.com/p/the-collapse-of-attribution), checked 8 October 2026
3. [support.google.com, GA4 Understand (direct) / (none) traffic, Analytics Help (opens in a new tab)](https://support.google.com/analytics/answer/15258820?hl=en), checked 8 October 2026
4. [graphite.io, Last-Touch Attribution Only Captures 10% of n8n's AEO Conversions, Five Percent (opens in a new tab)](https://graphite.io/five-percent/research/n8n-attribution-gap), checked 8 October 2026

AI can't recommend what it doesn't know.

[Check if AI recommends you](https://avouchedfor.com/assessment) 

Start free. No commitment.
