# Measuring Your Brand's Visibility in Claude Answers

> How to measure whether Claude mentions, recommends and cites your brand: a prompt set, clean test conditions, a scoring log and what to do with results.

Source: https://mentionx.ai/learn/claude/measure-brand-visibility-claude  
Author: Dipuck Jones, founder of MentionX  
Updated: 3 November 2026  
Level: Practitioner

**To measure your brand's visibility in Claude, ask it the questions your buyers ask, under clean and repeatable conditions, and log three things each time: whether you're mentioned, whether you're recommended and whether your site is cited.** One check tells you little, because answers vary between runs. A fixed set of questions, run several times on a regular schedule, shows you where you stand against competitors and whether your work is changing anything.

> **Key takeaways**
>
> - Measure mentions, recommendations and citations separately; they move for different reasons.
> - Test in incognito chats, which don't use or add to Claude's memory, so your own history doesn't colour answers.
> - When Claude searches the web, every answer includes citations, so you can see which pages it relied on.
> - Run each question at least three times and report patterns, not single answers.
> - Analytics shows only part of the picture: some Claude-driven visits arrive as claude.ai referrals and some carry no referrer at all.

*Checked 3 November 2026 (CET) against Anthropic help and docs pages. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration; its numbers are illustrative.*

## What exactly should you measure?

"Visibility" is too vague to act on. Break it into measures you can log in a sheet:

| Measure | Question it answers | How to log it |
| --- | --- | --- |
| Mention | Does Claude name us at all? | Yes or no |
| Recommendation | Does Claude suggest us as an option to choose? | Yes, no, or mentioned with a caveat |
| Position | Where do we appear in a list? | 1, 2, 3 and so on, or blank |
| Citation | Does Claude link to our pages? | Yes or no, plus which URL |
| Accuracy | Is what Claude says about us right? | Correct, outdated or wrong, with a note |
| Competitors | Who else appears? | Names, logged internally as Competitor A, B and so on |

Mentions without citations usually mean Claude knows your brand from its training or from other sites talking about you. Citations mean your own pages were found and used during a web search. The two respond to different work, which the previous lesson on [how Claude chooses sources](https://mentionx.ai/learn/claude/how-claude-chooses-sources-get-cited) explains.

*Our view:* accuracy is the most underrated measure. Being mentioned with an outdated price or a feature you dropped can do more harm than not being mentioned.

## How do you build the question set?

Use the questions buyers would actually type, not your keywords. Aim for 20 to 40 questions across the stages of a decision:

- **Problem:** "How do small finance teams stop paying duplicate invoices?"
- **Options:** "What tools automate accounts payable for a 100-person company?"
- **Comparison:** "[Your brand] vs [competitor] for invoice approvals"
- **Decision:** "Is [your brand] worth it for a team that uses Xero?"
- **Brand:** "What is [your brand]?" and "What does [your brand] cost?"

Problem and options questions matter most, because that's where buyers form a shortlist before they know your name. Brand questions are mainly for checking accuracy.

Claude can draft the list for you. Run this in a normal chat, then edit the result:

```
I sell [product] to [role] at [company type]. List 30
questions these buyers might ask an AI assistant while
choosing a [category] tool, grouped by stage: problem,
options, comparison, decision. Write them the way a buyer
would type them, in plain language, without our brand name
except in the comparison and decision groups.
```

The [prompt set guide](https://mentionx.ai/learn/measurement/building-prompt-set-many-prompts-need) goes deeper on how many questions you need and how to weight them.

## How do you set up clean test conditions?

Your own Claude account knows a lot about you, which can skew results. Control for it:

**Use incognito chats.** Incognito chats aren't saved to history or memory, and Claude doesn't use your memory in them ([Anthropic](https://support.claude.com/en/articles/12260368-use-incognito-chats)). That's the closest you'll get to what a new user sees. If you can't use incognito, turn memory off for the chat from the "+" menu ([Anthropic](https://support.claude.com/en/articles/11817273-using-claude-s-chat-search-and-memory-to-build-on-previous-context)).

**Stay out of projects.** Projects hold your brand documents and instructions. Testing inside one tells you what Claude says when it already has your files.

**Note the search setting.** Depending on your Claude experience, web search is either a toggle you turn on or something Claude does automatically when it helps ([Anthropic](https://support.claude.com/en/articles/10684626-enable-and-use-web-search)). Record whether each answer used search, because answers with and without it can differ a lot.

**Record the model and date.** Write down which model was selected and the date. When results change, you'll want to know if the model changed too.

**Paste the question exactly.** Don't rephrase between runs. Small wording changes can change answers.

## How many times should you run each question?

At least three times per round, in separate incognito chats. Claude generates each answer fresh, and when it searches, the sources it finds can change from one run to the next. A brand that appears in two of three runs is in a very different position from one that appears in three of three, and a single check can't tell them apart.

Log results as a rate: "mentioned in 7 of 9 runs across three questions" is more honest than "Claude recommends us". The guide to [week-over-week swings](https://mentionx.ai/learn/measurement/interpreting-week-over-week-visibility-swings) explains how much movement to expect before you treat a change as real.

> **Pro move:** when Claude cites sources, open the cited pages and log them, including the ones that aren't yours. The third-party pages Claude keeps citing for your category (review sites, comparison articles, community threads) are your outreach and PR list. Getting accurately described on those pages often moves mentions faster than changing your own site.

## How do you log and score results?

Keep one row per run in a sheet. These columns are enough:

```
date | model | search used (y/n) | question | stage | run #
| mentioned (y/n) | recommended (y/n/caveat) | position
| cited (y/n) | cited URL | accuracy (ok/outdated/wrong)
| competitors named | notes
```

Then report four numbers per round: mention rate, recommendation rate, citation rate and the share of answers with an accuracy problem. Break them down by stage, because "we're visible for brand questions but missing from options questions" is the most common and most useful finding.

To speed up scoring, paste a batch of answers into a normal chat and let Claude fill the sheet, then spot-check:

```
Below are [n] answers from Claude to buyer questions, each
labelled with the question and run number. For each answer,
fill these columns: mentioned [brand] (y/n), recommended
(y/n/caveat), position in any list, cited a [domain] URL
(y/n, which URL), competitors named, and any statement about
[brand] that contradicts this fact sheet: [paste].
Output a table only. If unsure about any cell, write "?".
```

Check every "?" and a sample of the rest yourself. Scoring errors add up quickly across hundreds of rows.

## What can analytics tell you?

Some people click links in Claude's answers, and those visits can show up in analytics as referrals from claude.ai. Not all do: some AI-driven visits arrive without referrer data and land in direct traffic, and many people read an answer and never click at all. Treat analytics as a lower bound.

Two things help. A custom channel group that collects AI referrers, including claude.ai, makes the traffic visible in one place; the [GA4 channel group guide](https://mentionx.ai/learn/measurement/ga4-custom-channel-group-ai-referrals) shows how. And adding "AI assistant (ChatGPT, Claude, etc.)" as an option on "How did you hear about us?" forms catches influence that analytics misses; see [self-reported attribution](https://mentionx.ai/learn/measurement/self-reported-attribution-adding-chatgpt-forms).

## When should you move from manual checks to tracking?

Manual checks are right for a first audit and for investigating a specific question. They become hard to sustain past about 20 questions run weekly, and they can't easily show trends across engines. At that point, an automated tracker that runs the same prompts on a schedule and stores the answers saves time and gives you comparable history. The guide to [manual and automated tracking](https://mentionx.ai/learn/measurement/manual-automated-ai-visibility-tracking) covers the trade-offs.

Whichever you use, keep a small manual check every month. Reading full answers shows you things scores don't, such as the tone Claude uses about you or a competitor claim that keeps appearing.

## What do you do with the results?

Match each finding to an action:

| Finding | Likely cause | What to try |
| --- | --- | --- |
| Not mentioned for options questions | Few sources describe you as an option in the category | Get listed and described accurately on the comparison and review pages Claude cites |
| Mentioned but not cited | Claude knows you, but your pages aren't the best source found | Publish clear answer pages for those questions (lesson 19) |
| Wrong or outdated facts | Old pages, inconsistent descriptions across the web | Fix your own pages first, then the third-party pages Claude cites |
| Competitor recommended over you | Their strengths are better documented for that use case | Publish evidence for your strengths for that use case |

Re-run the same question set after changes. Give it several weeks, and judge the change by rates across runs, not one answer.

## Worked example: Ledgerline's first Claude audit

*Ledgerline is a fictional accounts-payable tool. Figures are illustrative.*

Ledgerline wrote 24 buyer questions and ran each three times in incognito chats, logging 72 answers. Brand questions looked fine: Claude described Ledgerline correctly in most runs. Options questions told a different story. Ledgerline was mentioned in fewer than a third of those runs, while Competitor A appeared in nearly all of them.

The cited sources explained why. Claude kept citing two comparison articles and a community thread that listed Competitor A and Competitor B but not Ledgerline. One brand answer also quoted an old starting price from a partner page.

The team asked the partner to update the price, published a page answering "how do small finance teams automate invoice approval" and contacted the authors of both comparison articles with current information. They scheduled the same 24 questions to run again in six weeks.

## What should you check before you finish the track?

- A question set of 20 to 40 buyer questions, grouped by stage.
- A clean test setup: incognito, outside projects, search use recorded.
- A log sheet with mention, recommendation, citation and accuracy columns.
- A first round of results, with at least three runs per question.
- One action for each of your biggest gaps.

You've reached the end of the track. Return to the [Claude for Marketers track](https://mentionx.ai/learn/claude/claude-for-marketers) to review the certificate project, or read how to [rank in Claude](https://mentionx.ai/learn/engines/claude/how-to-rank).

## Common questions

**How do I check if Claude mentions my brand?** Ask Claude the questions your buyers ask, in incognito chats with web search available, and log whether your brand is mentioned, recommended or cited. Repeat each question several times, because answers vary between runs.

**Why does Claude give different answers to the same question?** Responses are generated fresh each time, and when Claude searches the web the results it finds can change. Memory and project context can also shape answers, so test in clean conditions and look at patterns across runs.

**Can I see Claude traffic in Google Analytics?** Some visits from links in Claude answers can show up as referrals from claude.ai, but not all AI-driven visits carry referrer data. Treat analytics as a partial signal alongside your answer tracking.

## Sources

- Anthropic Help Center, [Use incognito chats](https://support.claude.com/en/articles/12260368-use-incognito-chats)
- Anthropic Help Center, [Use Claude's chat search and memory to build on previous context](https://support.claude.com/en/articles/11817273-using-claude-s-chat-search-and-memory-to-build-on-previous-context)
- Anthropic Help Center, [Enable and use web search](https://support.claude.com/en/articles/10684626-enable-and-use-web-search)
All accessed 3 Nov 2026.

## Frequently asked questions

### How do I check if Claude mentions my brand?

Ask Claude the questions your buyers ask, in incognito chats with web search available, and log whether your brand is mentioned, recommended or cited. Repeat each question several times, because answers vary between runs.

### Why does Claude give different answers to the same question?

Responses are generated fresh each time, and when Claude searches the web the results it finds can change. Memory and project context can also shape answers, so test in clean conditions and look at patterns across runs.

### Can I see Claude traffic in Google Analytics?

Some visits from links in Claude answers can show up as referrals from claude.ai, but not all AI-driven visits carry referrer data. Treat analytics as a partial signal alongside your answer tracking.

---

Written and reviewed by Dipuck Jones for MentionX Learn. The HTML version of this page: https://mentionx.ai/learn/claude/measure-brand-visibility-claude