# How to Track Brand Mentions in AI Search: The Manual Method, the Tools, and the Revenue Layer (2026)

> Track brand mentions in AI search across ChatGPT, Perplexity, Gemini, Claude, and Copilot: a 15-prompt manual method, verified tool pricing, and revenue math.

Author: Vincent Ruan (https://x.com/0xVinceAI)
Publisher: Attrifast — https://attrifast.com
Last updated: 2026-08-01
Canonical: https://attrifast.com/blog/track-brand-mentions-ai-search

*Part of the [AI revenue attribution hub](/ai-revenue-attribution) and the [AI visibility to revenue attribution pillar](/ai-visibility-to-revenue-attribution).*

**TL;DR**

- **Traditional brand monitoring is structurally blind here.** Google Alerts, Brand24, and Mention crawl published web pages. A ChatGPT answer is generated privately, per user, per session — it never becomes a page. If ChatGPT told 500 prospects this week that your competitor is the better pick, no alert fired anywhere.
- **The manual method works up to about 20 prompts:** 10–20 buying-intent prompts, run across the engines your buyers use, 2–3 phrasings each, logged in a spreadsheet. At 20 prompts × 5 engines × 3 variations that is 300 answer-checks per weekly run — roughly 7–8 hours. That is where tooling starts to pay.
- **Monitoring platforms, verified July 2026:** Otterly from $29/mo [5], Peec from €89/mo [6], Scrunch from $250/mo billed annually [7], Profound from $99/mo billed annually for a ChatGPT-only tier [8].
- **Mentions are not money.** 73% of ChatGPT-referred sessions land in GA4's Direct bucket [2][10], and revenue per visitor ranges from $0.29 (AI Overviews) to $1.42 (Perplexity) [10]. Attrifast does not do prompt-level monitoring — it tells you which engine's mentions become Stripe revenue. The honest stack is one monitoring tool for optimization plus [revenue attribution](/ai-revenue-attribution) for the dollars → [Start free trial](/login)

Somewhere in the last two years, "brand monitoring" quietly split into two different problems. The old problem — who is writing about us on the public web — is solved, cheap, and automated. The new problem is that the most influential brand mentions now happen inside AI answers: ChatGPT has roughly 800 million weekly users [1], and when one of them asks "what's the best invoicing tool for freelancers," the answer names three brands, describes each in a sentence, and disappears when the session ends.

No crawler saw it. No alert fired. And the referrer usually dies on the click, so even your analytics tool files the visit under Direct [2].

This guide is the hands-on answer to how to track brand mentions in AI search in 2026: the manual method that costs nothing but time, the point where it mathematically breaks, the tool landscape with verified pricing, and the layer almost everyone skips — connecting mentions to revenue. If you want a buyer's guide to the tooling category itself, that's [our AI visibility tools guide](/blog/ai-visibility-tools-guide); this article is the how-to.

## Why AI brand monitoring needs a different playbook

Three properties make AI-answer mentions unlike anything Google Alerts was built for:

**1. Answers are ephemeral and private.** A web mention is a URL — crawlable, linkable, permanent-ish. An AI mention exists for one user, in one session, and is gone. The only way to observe it is to *ask the engine yourself* and treat the sampled answers as a survey of what real users see.

**2. Answers are probabilistic.** Ask ChatGPT the same question five times and you will get materially different brand lists. Wording, session context, and model updates all shift results. One check is an anecdote; a mention *rate* across repeated, varied runs is a measurement. The Princeton GEO study formalized this — measuring visibility across large prompt sets, and showing optimization can shift answer visibility by up to 40% [4].

**3. The click evidence gets destroyed.** Even when a mention drives a visit, the engines frequently strip the referrer. Independent measurement puts 65–82% of ChatGPT-originated visits in GA4's Direct bucket [2], and our own 200-site data says 73% [10]. So you cannot back into mention performance from analytics alone, either.

Put together: you cannot crawl the mentions, you cannot trust a single check, and you cannot see most of the clicks. The playbook below deals with each in turn.

## How to track brand mentions in AI search manually

The manual method is genuinely good — it is how we ran [our own first visibility scan](/blog/analyze-competitor-ai-visibility) before automating anything. It has five steps.

### Step 1: Build a buying-intent prompt set (10–20 prompts)

Do not monitor prompts about your brand ("what is Attrifast?") — people who already know your name are not the prize. Monitor the prompts where buyers *discover* brands:

- **Category prompts:** "best [category] tools 2026", "top [category] for small business"
- **Use-case prompts:** "how do I [job to be done]", "[category] for [specific ICP]"
- **Comparison prompts:** "[competitor A] vs [competitor B]", "cheaper alternative to [market leader]"
- **Problem prompts:** "[pain point] — what should I use?"

Ten to twenty prompts is the right size. Fewer misses whole intent categories; more breaks the weekly arithmetic (Step 5). Pull phrasing from your sales calls, support tickets, and the People Also Ask boxes on your money keywords.

### Step 2: Pick your engines — in order of where your buyers are

Cover ChatGPT first; it is not close. Across the 200 Stripe-connected sites in our benchmark, ChatGPT carries 71% of AI-referred sessions:

**AI-engine session share — where to spend your manual checking time (% of AI sessions)**

| Category | Value |
|---|---|
| ChatGPT | 71% |
| Gemini | 12% |
| Perplexity | 8% |
| Claude | 6% |
| AI Overviews | 3% |

*Source: Attrifast 200-site benchmark, per-engine session share, 2026*

The practical order for most B2B and SaaS brands: **ChatGPT, then Gemini (which shares plumbing with Google AI Overviews and AI Mode), then Perplexity, then Claude**, with AI Overviews and Copilot checked monthly rather than weekly. Note the trap in this chart: session share is not revenue share — Perplexity and Claude send far fewer visitors who are worth far more each [10]. More on that in the revenue section.

### Step 3: Run each prompt with 2–3 variations

Because answers are probabilistic, a single run tells you almost nothing. Run each prompt in a fresh session (no chat history, logged out or in a temporary chat where possible), then repeat with two paraphrases:

- "best accounting software for freelancers"
- "what accounting tool should a freelancer use?"
- "I'm a freelancer, recommend accounting software"

If your brand appears in 7 of 9 runs, you have a mention rate of ~78% for that intent. If it appears in 2 of 9, you learned something a single lucky check would have hidden.

### Step 4: Log every answer in a fixed template

A spreadsheet with one row per answer-check is enough. The columns that matter:

| Column | What to record |
| --- | --- |
| Date / Engine / Prompt / Variation # | The check's coordinates |
| Mentioned? | Y/N — your brand named anywhere in the answer |
| Position | 1st, 2nd, 3rd… in the recommendation list |
| Sentiment | Positive / neutral / negative, plus the *exact descriptor words* |
| Cited URL | Which page of yours (or of a third party) the engine linked |
| Competitors named | Every other brand in the answer |

Two of these columns earn their keep disproportionately. **Descriptor words** are your LLM brand sentiment signal: engines reuse phrases like "affordable but limited" across thousands of sessions, so one recurring negative descriptor compounds like no single bad review ever did — we covered the revenue side of this in [our AI brand sentiment analysis](/blog/ai-brand-sentiment-revenue-impact). **Competitors named** turns your log into a share-of-voice tracker for free; the full competitor-teardown method is in [how to analyze competitors' AI visibility](/blog/analyze-competitor-ai-visibility).

### Step 5: Repeat weekly — and know where the arithmetic breaks

Weekly is the floor. AI answers move faster than rankings: engines re-retrieve sources, models update, and citation sets churn. But here is the workload math nobody puts on the pricing page:

**Manual answer-checks per weekly run: prompts × 5 engines × 3 variations**

| Category | Value |
|---|---|
| 10 prompts | 150 checks |
| 15 prompts | 225 checks |
| 20 prompts | 300 checks |
| 30 prompts | 450 checks |
| 50 prompts | 750 checks |

*Source: Workload arithmetic — prompt-set size × 5 engines × 3 paraphrase variations, one weekly run*

At 20 prompts, one weekly run is **300 answer-checks** — ask, read, log, next. At a realistic 90 seconds per check, that is seven and a half hours, every week, forever. At 50 prompts it is 750 checks and a part-time job. This is not a tooling sales pitch; it is why every team we know that started manual either capped their prompt set at ~15, dropped to monthly (and lost the trend line), or moved to a tool. The manual method's real job is to teach you *which 15 prompts matter* — then you decide whether to keep grinding or automate.

## The tool landscape: three tiers, honestly scoped

There are three ways to automate, and they are not interchangeable. (The full 12-tool comparison lives in [our AI visibility tools guide](/blog/ai-visibility-tools-guide); this is the decision-level view.)

### Tier 1: Mention-monitoring platforms

These do exactly what Steps 1–5 do, daily, at scale: run your prompt set across engines, log mentions, position, sentiment, and citations, and chart the trend. The four we consider credible in 2026, with pricing verified from each vendor's public page as of July 2026:

**AI mention-monitoring platforms — monthly price by tier (USD, as of July 2026)**

| Category | Value |
|---|---|
| Otterly Lite | $29/mo |
| Peec Starter (€89) | $95/mo |
| Profound Starter (annual) | $99/mo |
| Otterly Standard | $189/mo |
| Scrunch Starter (annual) | $250/mo |
| Profound Growth (annual) | $399/mo |

*Source: Verified from each vendor's public pricing page, July 2026*

| Tool | Entry price (July 2026) | What you get | Watch for |
| --- | --- | --- | --- |
| **Otterly.ai** [5] | $29/mo (Lite) | 15 prompts, daily runs across ChatGPT, AI Overviews, Perplexity, Copilot | Gemini and Claude are paid add-ons; 15 prompts is tight — the $189 Standard tier (100 prompts) is the realistic plan |
| **Peec AI** [6] | €89/mo (~$95, Starter) | Six engines on every plan (ChatGPT, AI Mode, AI Overviews, Copilot, Perplexity, Gemini), unlimited seats | Priced in EUR; prompt allowances scale with tier |
| **Profound** [8] | $99/mo billed annually (Starter) | 50 prompts — but ChatGPT only at this tier | Multi-engine coverage starts at Growth, $399/mo billed annually; built for mid-market and up |
| **Scrunch** [7] | $250/mo billed annually ($300 monthly) | 350 custom prompts, personas, page audits | Enterprise-leaning; overkill below ~$1M ARR |

All four are good at the monitoring job. Notice what none of them list on any tier: revenue. A mention dashboard can tell you visibility went from 12% to 31%; it cannot tell you whether that was worth $40 or $40,000. That is not a criticism — it is a category boundary, and it is why the vendors themselves increasingly resort to signup surveys to estimate impact [3].

### Tier 2: DIY — API scripts plus a spreadsheet

If you have an engineer-hour to spare, the manual method automates directly: a script that hits the OpenAI, Anthropic, and Google APIs (and Perplexity's, which returns its citations) with your prompt set on a cron job, parses answers for brand names, and appends rows to a sheet or database. Cost at 20 prompts × 3 variations × 4 engines daily is typically single-digit dollars per month in API calls with cheap model tiers.

Two honest caveats. API responses are *not identical* to what consumer apps show — no user context, sometimes different retrieval — so treat DIY numbers as a directional index, not ground truth. And you will spend more time than you expect on parsing edge cases (brand nicknames, "Attrifast's" vs "Attrifast", competitors with dictionary-word names). It is the right tier for engineering-led teams who want to own their data; everyone else should buy Tier 1.

### Tier 3: The revenue layer — what Attrifast actually does (and does not do)

Full disclosure, since this is our blog: **Attrifast is not a prompt-level mention-monitoring tool.** It does not re-run 300 prompts a day, and if daily mention tracking is your primary need, buy one of the Tier 1 tools above.

What Attrifast does is the layer the monitoring tier stops at:

- A **monthly [AI Visibility scan](/features/ai-visibility-score)** that samples the four major engines with category and commercial prompts and reports whether you were mentioned and cited — a pulse check, not a daily monitor, and we say so plainly.
- **Server-side detection of AI-referred visits joined to Stripe payments** — the actual product. Every visit from ChatGPT, Perplexity, Claude, Gemini, Copilot, or AI Overviews is detected at the edge (including most of the ones whose referrer was stripped) and joined to the payment webhook, producing per-engine visits, conversion rate, and revenue. That is the [AI revenue attribution](/ai-revenue-attribution) job, at $9.99/mo.

The pairing we genuinely recommend — including to our own customers — is one Tier 1 tool to optimize mentions, plus Attrifast to verify the money. One tells you *where you appear*; the other tells you *what appearing is worth*. Neither substitutes for the other.

## From mentions to money: why mention counts alone mislead

Here is the funnel every mention travels: **mentioned → cited → clicked → converted.** Monitoring tools measure the first two stages. Revenue lives at the end. And the pipe between them leaks in a way that flatters mention counts.

Start with the click evidence. When an AI mention does drive a visit, the referrer usually dies in transit:

**Where ChatGPT-referred sessions land in GA4 (% of visits)**

| Category | Value |
|---|---|
| Misattributed to Direct/(none) | 73% |
| Correctly attributed to AI engine | 21% |
| Bucketed as Referral/Other | 6% |

*Source: Attrifast measurement of AI referral attribution in GA4, Q1–Q2 2026*

73% of ChatGPT-referred sessions land in Direct/(none); only 21% are correctly attributed [10] — consistent with SE Ranking's independent 65–82% range [2]. You can recover a slice with a GA4 custom channel group [9], and you should. But the structural picture, per thousand AI-referred visits, looks like this:

**Per 1,000 AI-referred visits: what analytics sees vs what converts**

| Category | Value |
|---|---|
| AI visits actually arriving | 1000 |
| Labeled correctly as AI | 360 |
| Converting to paid customers | 27 |

*Source: Attrifast 200-site Stripe-connected benchmark, 2025-06 to 2026-05*

A thousand visits arrive; a default analytics setup labels about 360 of them correctly [10]; roughly 27 convert — because AI-referred visitors convert at 2.7% against 1.4% for Google organic in the same cohort [10]. The conversions come from all 1,000 visits, but your dashboard only connects a third of them to AI. Which means any ROI math built on "mentions went up, and analytics shows some AI traffic" is running on a third of the evidence.

The second distortion is that engines are not worth the same per mention:

**Revenue per visitor by AI engine (USD)**

| Category | Value |
|---|---|
| Perplexity | $1.42 |
| Claude | $1.18 |
| ChatGPT | $0.87 |
| Gemini | $0.41 |
| AI Overviews | $0.29 |

*Source: Attrifast 200-site benchmark, cohort-blended RPV, 2025-06 to 2026-05*

Perplexity sends 8% of AI sessions and $1.42 per visitor; AI Overviews sends 3% of sessions at $0.29 [10]. A mention dashboard that weights all engines equally will systematically steer your optimization effort toward volume and away from value. Whether GEO work pays back at all is a longer argument — we ran the numbers in [does GEO actually drive revenue](/blog/does-geo-actually-drive-revenue), and the measurement architecture for connecting the two layers is the subject of [the AI visibility to revenue attribution pillar](/ai-visibility-to-revenue-attribution). The mechanics of catching the stripped-referrer traffic specifically are in [the ChatGPT traffic tracking guide](/track-chatgpt-traffic).

The takeaway is not "ignore mentions." Mentions are the leading indicator and the thing you can act on. The takeaway is: **never report mentions without a revenue column next to them**, or you will one day defend a budget with a number that a CFO can dissolve in one question.

## FAQ: tracking brand mentions in AI search

### Can Google Alerts track brand mentions in ChatGPT?

No. Google Alerts, Brand24, Mention, and every traditional tool work by crawling published web pages. A ChatGPT answer is generated privately for one user in one session and never becomes a crawlable page — the same is true for Claude, Gemini, and Copilot. The only ways to observe AI-answer mentions are to ask the engines yourself (manually or via API) or use a monitoring platform that does it at scale.

### How do I track brand mentions in AI search for free?

Run the manual method above: 10–20 buying-intent prompts, run in ChatGPT, Perplexity, Gemini, and Claude with 2–3 phrasings each, logged weekly in a spreadsheet. It costs nothing but time — about 7–8 hours per weekly run at the 20-prompt scale. Past that, a $29–95/month tool is cheaper than your hours.

### What are the best tools to track mentions in ChatGPT?

As of July 2026: Otterly.ai from $29/mo (15 prompts, four engines) [5], Peec AI from €89/mo (six engines on every plan) [6], Scrunch from $250/mo billed annually [7], and Profound from $99/mo billed annually for its ChatGPT-only tier [8]. None tracks revenue — pair one with payment-verified attribution if you need to know what the mentions are worth.

### How often should I check my brand mentions in AI engines?

Weekly is the practical floor for manual checks; monitoring platforms re-run prompts daily. AI answers churn faster than search rankings because engines re-retrieve sources and vary responses across sessions. Whatever cadence you choose, keep it fixed — mention rates measured on an inconsistent schedule cannot show trend, and trend is the point.

### How do I monitor brand mentions in Gemini?

Manually, run your prompt set in the Gemini app and in Google AI Mode and log results like any other engine. Via tools: Peec includes Gemini on every plan [6]; Otterly offers it as a paid add-on [5]; Profound covers it from the Growth tier [8]. Gemini brand monitoring matters beyond its 12% session share because it shares infrastructure with AI Overviews and AI Mode — visibility gains there tend to travel across all three Google surfaces.

### What is LLM brand sentiment, and how do I measure it?

LLM brand sentiment is how an engine *describes* you when it mentions you — recommended enthusiastically, listed neutrally, or hedged with "expensive," "dated," "limited support." Measure it by logging a sentiment grade plus the exact descriptor words on every check. Descriptors are the leading indicator: engines reuse them across thousands of sessions, so a single recurring negative phrase compounds. We quantified the revenue impact in [our AI brand sentiment breakdown](/blog/ai-brand-sentiment-revenue-impact).

### Does Attrifast track brand mentions in ChatGPT?

Not at the prompt level. Attrifast runs a monthly [AI Visibility scan](/features/ai-visibility-score) — a sampled pulse check across four engines — and its core product is server-side detection of AI-referred visits joined to Stripe payments: per-engine visits, conversion, and revenue. For daily prompt-level monitoring, use a Tier 1 tool; for knowing which engine's mentions become dollars, that is [what Attrifast is for](/ai-revenue-attribution).

## The bottom line

Tracking brand mentions in AI search is three jobs wearing one name. **Sampling** what the engines say — start manual with 15 buying-intent prompts, move to a tool when the arithmetic breaks. **Monitoring** the trend — Otterly, Peec, Scrunch, or Profound, depending on your prompt count and budget, all verified above. And **valuing** the mentions — the job the monitoring category structurally cannot do, because the referrer dies before the dashboard and the dashboard never sees the payment.

Do the first two and you will know where you stand in AI answers. Add the third and you will know whether standing there matters. Attrifast does exactly one of these jobs — the [revenue one](/ai-revenue-attribution) — and it takes about two minutes to wire up: one 4kb script, one Stripe connection, $9.99/mo. [Start the free trial](/login) and find out what your mentions have been quietly worth.

## Sources

1. [ChatGPT traffic tracker — AI chatbot referral and usage volumes](https://aisearch.similarweb.com/ai-chatbot-traffic/chatgpt-traffic-tracker/) — Similarweb, 2026
2. [ChatGPT referral traffic study — May 2026 measurement of AI referral patterns](https://seranking.com/blog/chatgpt-referral-traffic-may-2026/) — SE Ranking, 2026
3. [Claude Drives 10.6% of Our Signups. Google Analytics Says 0.1% — measuring AI search conversions](https://otterly.ai/blog/measure-ai-search-conversions/) — Otterly.ai, 2026
4. [GEO: Generative Engine Optimization — visibility measurement across generative engines](https://arxiv.org/abs/2311.09735) — arXiv (Aggarwal et al., Princeton/Georgia Tech), 2024
5. [Otterly.ai pricing — Lite $29/mo (15 prompts), Standard $189/mo, Premium $489/mo](https://otterly.ai/pricing/) — Otterly.ai, 2026
6. [Peec AI pricing — Starter €89/mo, six engines on all plans](https://peec.ai/pricing) — Peec AI, 2026
7. [Scrunch pricing — Starter $250/mo billed annually ($300 monthly)](https://scrunch.com/pricing/) — Scrunch, 2026
8. [Profound pricing — Starter $99/mo and Growth $399/mo, billed yearly](https://www.tryprofound.com/pricing) — Profound, 2026
9. [How to track AI traffic in Google Analytics 4 — custom channel group setup](https://www.analyticsmania.com/post/ai-traffic-in-google-analytics-4/) — Analytics Mania, 2026
10. [AI Traffic Revenue Benchmark 2026 — 200 Stripe-connected sites, per-engine RPV and conversion](https://attrifast.com/blog/ai-traffic-revenue-benchmark-2026) — Attrifast, 2026