How to Track Your Brand in AI Answers (Step-by-Step)

Brand tracking in AI search requires a fundamentally different methodology from traditional rank tracking. The same query returns different brand recommendations on approximately 99 of 100 runs. Single-query checks are noise. Reliable tracking requires multi-run aggregation across a structured query set.

This guide walks through the 5-step methodology we use to track brand visibility across ChatGPT Search, Perplexity, Google AI Overviews, and Claude. By the end, you will have a repeatable weekly process that surfaces real trends rather than single-run volatility.

Step 1: Build the query set

The query set is the foundation. A bad query set produces bad data regardless of how carefully you run it. Include four query types in roughly equal proportions:

Aim for 50–200 queries total. Under 50, the data is too sparse. Over 200, the cost and time become prohibitive. 100 queries is a strong starting point — 25 of each type.

Step 2: Run multi-pass

This is the step most brands skip — and the reason most brand tracking is noise. Run every query 10+ times per engine. ChatGPT, Perplexity, Google AI Overviews, and Claude all generate answers probabilistically. Single-run results reflect randomness, not reality.

"The same query, run 100 times across major AI search engines, returns different brand recommendation lists on approximately 99 of those runs. Single-run rank tracking is noise. Multi-run aggregation is signal."

Minimum: 10 runs per query per engine. Recommended: 25 runs. For 100 queries × 10 runs × 3 engines, that is 3,000 query executions per week. Manual execution is impossible at this volume — use a tracking tool (see our 7 Best AI Search Visibility Tools for GEO Tracking (2025)) or build a scripted pipeline.

Step 3: Extract mentions, citations, and sentiment

For each answer, extract three signals:

1. 1.
   Mention

   Does your brand name appear anywhere in the answer? Log as binary (yes/no) per run.

2. 2.
   Citation

   Is your content cited as a source? Log as binary per run. Track which URL is cited.

3. 3.
   Sentiment

   Is the framing positive, neutral, or negative? Use an LLM-based classifier for consistency.

Optional fourth signal: competitor mentions. Track which competitors appear in the same answers as you, for share-of-voice calculations.

Step 4: Aggregate per query and per engine

Aggregation converts 10 runs of one query into a single reliable data point. Compute three aggregates per query per engine:

• ▸ Mention rate = (runs where brand appears ÷ total runs) × 100
• ▸ Citation rate = (runs where brand is cited ÷ total runs) × 100
• ▸ Positive sentiment share = (positive mentions ÷ total mentions) × 100

Then average across queries of the same type (branded, category, comparison, informational) and across all queries. The result: a single weekly snapshot per engine, broken down by query type.

Step 5: Compare week-over-week and trend

Single-week numbers are still noisy even with 10-run aggregation. The signal emerges in trends. Compare each week's snapshot to the previous week, the previous month, and the previous quarter.

React to 4-week trends, not single-week swings. A 5-point drop in mention rate over one week is volatility. A 5-point drop sustained over 4 weeks is a real problem requiring investigation.

Engine-specific notes

Each AI engine has different citation patterns. Tracking methodology must adjust:

Source: OpenAI, Perplexity, Google, Anthropic documentation (2025). Seer Interactive Google AIO coverage study (2025).

Common tracking mistakes

• ▸ Single-run tracking — One query execution per week. Pure noise.
• ▸ Branded queries only — Misses category, comparison, and informational visibility.
• ▸ One engine only — Each engine has different audiences and citation patterns.
• ▸ Daily cadence — Day-to-day swings are volatility, not signal.
• ▸ No sentiment — High mention rate with negative sentiment is a problem.
• ▸ Reacting to weekly swings — Wait for 4-week trends before acting.

Frequently asked questions