Definition
Share of Model tracking is the ongoing practice of measuring how often a brand appears in answers from large language models, compared with its competitors, and recording how that changes over time. It involves running a fixed set of prompts through AI systems such as ChatGPT, Gemini, Claude and Perplexity on a regular schedule and logging which brands each answer names and which sources it cites.
Share of Model (SoM) is the metric. Share of Model tracking is the program that produces it: the prompt set, the schedule, the log and the reporting.
The practice also goes by other names. Tool vendors call it AI visibility tracking, LLM visibility tracking, prompt tracking or AI brand monitoring. The methods are largely the same.
See also: Share of Model (SoM), Generative Engine Optimization (GEO), Share of Voice (SOV), Citation Optimization
Why it matters for marketing
A single check of what ChatGPT says about a brand tells a marketer very little. AI answers vary from one run to the next. They differ between platforms, and they shift when a model is updated, usually without notice. A brand that appears in an answer on Tuesday may be absent from the same question on Thursday.
Tracking deals with that by treating AI answers as something to sample repeatedly. Over weeks and months it shows whether a brand’s presence is rising or falling, which competitors are gaining, and whether content and PR work is changing anything.
It also gives teams a baseline. Without one, there’s no way to tell whether a generative engine optimization effort worked, or whether a drop in AI referral traffic came from a model change or from a competitor’s new content.
For marketing leaders, tracking turns a vague concern (“are we showing up in AI?”) into a reportable number with a trend line.
Hear it on The Agile Brand
These episodes of The Agile Brand with Greg Kihlström cover how brands are approaching AI search and visibility.
- #801: Brandlight CEO Imri Marcus on GEO, AI browsers and the new search experience
- #793: PAN Communications CMO Mark Nardone on 2026 predictions and staying ahead of GEO
Find more episodes on The Agile Brand podcast page.
How it works
A tracking program has five parts.
| Component | What it is | Decisions to make |
|---|---|---|
| Prompt set | The fixed list of questions run each cycle | Which buyer questions, how many, branded or unbranded |
| Competitor set | The brands counted alongside yours | Which competitors, and how name variants are handled |
| Platforms | The AI systems queried | Which ones your buyers use |
| Cadence and sampling | How often prompts run and how many times each | Weekly or monthly, and the number of repeat runs |
| Log | The record of each answer | What fields to capture |
Building the prompt set
The prompt set is the measuring instrument, so the result is only as good as the questions in it. Useful prompts come from real demand: search query data, sales call notes, customer support questions and the questions prospects ask in demos.
Most programs group prompts by type.
- Category prompts. “What are the best customer data platforms for retail?”
- Problem prompts. “How do I unify customer data across channels?”
- Comparison prompts. “Vendor A vs. Vendor B” or “alternatives to Vendor A.”
- Branded prompts. “What is Vendor A known for?”
Branded prompts are tracked separately. A prompt that contains the brand’s name will nearly always produce an answer that mentions it, which inflates the overall figure. The headline number should come from the unbranded set.
Once the set is chosen, it should stay fixed. Changing prompts between reporting periods breaks the comparison.
What to log
For each answer, a typical log records:
- The prompt, the platform and the date
- Whether the brand was mentioned
- Its position among the brands named
- Which competitors were mentioned
- Which domains and pages were cited as sources
- How the brand was described, including sentiment and accuracy
Recording the cited sources matters as much as recording the mentions. The sources show which sites are shaping the answer, and that’s where the work to change it usually starts.
How to calculate tracking volume and results
Two calculations come up in every program.
Answers collected per cycle = prompts × platforms × runs per prompt
An illustrative example: a B2B software company tracks 50 unbranded prompts across 4 AI platforms and runs each prompt 5 times per cycle. That’s 50 × 4 × 5 = 1,000 answers per cycle.
The volume grows quickly. One tracking vendor, LLM Pulse, notes that a prompt run weekly produces about 4.33 answers per model each month, so 50 prompts across 5 models is more than a thousand answers a month before any repeat runs.
Mention rate = (answers that mention the brand ÷ total answers) × 100
If the brand appears in 280 of those 1,000 answers, its mention rate is 28%.
The same log supports the competitive version, where the brand’s mentions are divided by total mentions of all tracked brands. That figure is commonly reported as Share of Model or AI share of voice.
How to set up Share of Model tracking
- Set the scope. Decide which product lines, markets and languages to cover.
- Build and freeze the prompt set. Start with a few dozen prompts tied to real buyer questions.
- Fix the competitor list. Adding a dominant competitor later lowers your share without anything real having changed, so set the list at the start and state it in every report.
- Choose platforms. Cover the AI systems your buyers use. Tracking one model says little about the others.
- Run a baseline. Collect the first full cycle before making any content changes.
- Set the cadence. Weekly is a common default for automated tools. Manual programs often run monthly.
- Report with context. Show the trend, the breakdown by platform and prompt type, and the most-cited sources.
- Act and re-measure. Use gaps to guide content and PR work, then watch the next cycles for movement.
Ways to run it
| Approach | How it works | Suits | Trade-offs |
|---|---|---|---|
| Manual | A person runs prompts and logs answers in a spreadsheet | Small teams, early exploration, 20 to 30 prompts | Slow, hard to sample enough, prone to inconsistency |
| AI visibility tool | Software runs the prompt set on a schedule and parses answers | Most ongoing programs | Subscription cost. Each tool uses its own method |
| SEO suite add-on | AI tracking inside an existing SEO platform | Teams already using that platform | Platform coverage and prompt sets vary |
| Agency or consultancy | An outside firm runs tracking and recommends actions | Teams without in-house capacity | Cost, and dependence on the firm’s method |
| Custom build | In-house scripts query AI systems through their APIs | Large programs with engineering support | Build and maintenance effort. API answers can differ from what consumers see in the apps |
Whatever the approach, the method should be written down: the prompts, the competitors, the platforms, the number of runs and the formula. Two programs using different methods will produce different numbers for the same brand.
Comparison with related practices
| Practice | What it tracks | Where | Output |
|---|---|---|---|
| Share of Model tracking | Brand mentions and citations in AI answers over time | AI assistants and AI search | Trend in Share of Model |
| Keyword rank tracking | A page’s position for search queries | Search engine results | Ranking positions |
| Media monitoring | Brand mentions in news, social and broadcast | Earned and social media | Mention volume, sentiment |
| Share of voice measurement | Brand’s portion of ad or media presence | Paid, owned and earned media | Share of voice |
| AI referral tracking | Visits arriving from AI platforms | Web analytics | Sessions and conversions |
| Brand tracking survey | Awareness and perception among people | Survey panels | Awareness, consideration |
Rank tracking is the closest comparison and the most misleading one. A search ranking is a fixed position that everyone sees. An AI answer is generated fresh each time and can differ between users, so tracking measures a rate across many samples instead of a position.
Share of Model tracking and AI referral tracking work as a pair. The first shows whether the brand is appearing in answers. The second shows whether those appearances send anyone to the site.
Best practices
Freeze the prompt set and competitor list. Review them on a schedule, such as quarterly, and note any change in reports.
Lead with unbranded prompts. Report branded prompts separately.
Sample each prompt more than once. One answer per prompt isn’t a stable reading.
Track several platforms. A brand can be strong in one model and absent from another.
Log sources, not only mentions. Cited domains show what to work on.
Check accuracy. Record whether the AI describes the brand, its products and its pricing correctly.
Annotate model updates. Note major model releases on the trend line so shifts can be explained.
Don’t compare across tools or prompt sets. The numbers aren’t equivalent.
Pair it with business metrics. Report AI referral traffic and pipeline next to the share figure.
Assign an owner. Tracking usually sits with SEO, content or brand, and it needs one accountable person.
Limitations
Sampled, not observed. Tracking shows what AI systems say in response to test prompts. It doesn’t show what real users asked or saw.
No demand data. AI platforms don’t publish how often a given question is asked, so every prompt is usually weighted the same or by estimate.
Personalization and location. Answers can vary by user history and geography in ways a test doesn’t reproduce.
App and API differences. Answers gathered through an API may not match the consumer product.
Unannounced model changes. A shift in the numbers may reflect a model update, with no change in the brand’s own activity.
Method differences between vendors. There’s no standard, so figures aren’t portable.
Future trends
First-party data from AI platforms. If AI providers begin reporting brand impressions directly, tracking will rely less on sampling.
Agent tracking. As AI agents browse and buy on behalf of users, tracking is extending from what models say to what agents select.
Larger prompt panels. Tool vendors are moving from hundreds of prompts to very large panels to reduce sampling error.
Consolidation into existing suites. AI tracking is being folded into SEO and analytics platforms that marketing teams already use.
Paid placements. Advertising inside AI answers will require tracking to separate paid presence from earned.
Standardization efforts. Demand for consistent definitions is growing as the figures reach executive reporting.
Frequently asked questions
What is Share of Model tracking? It’s the practice of running a fixed set of prompts through AI systems on a schedule and recording how often a brand and its competitors appear in the answers.
How is it different from Share of Model? Share of Model is the metric. Tracking is the repeated measurement process that produces the metric and shows its trend.
How often should Share of Model be tracked? Weekly is a common default for automated tools, and monthly is typical for manual programs. Priority prompts are sometimes run more often than the rest.
How many prompts are needed? There’s no fixed number. Manual programs often start with 20 to 30. Tool-based programs commonly use 50 to a few hundred.
Why run the same prompt several times? AI answers aren’t identical from run to run. Repeating each prompt gives a more stable rate.
Which AI platforms should be tracked? The ones your buyers use. Programs commonly include ChatGPT, Gemini, Perplexity, Claude and Google’s AI Overviews or AI Mode.
Can Share of Model be tracked manually? Yes, with a spreadsheet and a small prompt set. It becomes impractical as prompts, platforms and repeat runs increase.
What tools track Share of Model? Specialized AI visibility tools and several SEO suites offer it. Each uses its own prompts and formulas, so results differ between them.
Should a company use an agency for Share of Model tracking? It depends on capacity. An agency can run the program and recommend actions. A team with SEO or analytics staff can often run it in-house with a tool. In both cases, ask for the method in writing.
What should a tracking report include? The overall share and its trend, a breakdown by platform and prompt type, competitor comparison, the most-cited sources, any accuracy problems and notes on model changes.
Related terms
- Share of Model (SoM)
- Share of Voice (SOV)
- Generative Engine Optimization (GEO)
- Answer Engine Optimization (AEO)
- Citation Optimization
- AI Overviews
- Large Language Models (LLM)
- llms.txt
- Key Performance Indicator (KPI)
- Search Engine Optimization (SEO)
Sources
- LLM Pulse. “How to Track Brand Mentions in ChatGPT (2026 Guide).” https://llmpulse.ai/blog/track-mentions-chatgpt/
- LLM Pulse. “Prompt Tracking” (glossary). https://llmpulse.ai/blog/glossary/prompt-tracking/
- AirOps. “LLM Visibility” (glossary). https://www.airops.com/glossary/llm-visibility
- Sight AI. “How to Monitor LLM Responses.” https://www.trysight.ai/blog/how-to-monitor-llm-responses
- Sight AI. “How to Track LLM Brand References: A Step-by-Step Guide for Marketers and Founders.” https://www.trysight.ai/blog/track-llm-brand-references
- Senso. “The Complete Guide to Tracking Your Brand in ChatGPT, Claude, and Perplexity.” https://senso.ai/prompts-content/guide-to-tracking-brand-mentions-in-chatgpt-and-llms
- Metehan Yesilyurt. “AI Share of Voice Benchmarking Methods” (PDF). https://metehan.ai/pdf/ai-share-of-voice-benchmarking-methods.pdf
- Moburst. “How to Track Your Brand Across 5 AI Models.” August 13, 2026. https://answerburst.moburst.com/how-to-track-your-brand-across-5-ai-models/
- The Agile Brand Guide. “Share of Model (SoM).” https://www.agilebrandguide.com/wiki/agentic-commerce/share-of-model-som/