Buyers now ask AI assistants for recommendations before they open a search engine. If ChatGPT, Gemini, or Perplexity names your competitor and skips your brand, you lose the shortlist without seeing a single drop in your analytics.
Traditional rank tracking cannot show this problem. AI answers are generated fresh, change between sessions, and rarely follow a fixed order of results. That is why marketers now track two newer metrics: Share of Model and brand citation frequency.
What Share of Model and Citation Frequency Mean
Share of Model is the percentage of relevant AI answers in which your brand appears, compared with the other brands in your category. Think of it as market share, but measured inside AI responses instead of sales.
Brand citation frequency counts how often an AI platform links to or lists your website as a source for its answer. A mention is the AI saying your name. A citation is the AI pointing to your content as evidence.
The two are related but not identical. A brand can be mentioned often and never cited, or cited often without being recommended.
Metric
What it tells you
Example
Mention rate
How often your brand name appears in answers
Named in 30 of 100 answers
Share of Model
Your mentions compared with all tracked brands
30 of 120 total brand mentions
Citation frequency
How often your domain is shown as a source
Linked in 18 of 100 answers
Recommendation rate
How often you are the top or only pick
First named in 12 of 100 answers
Sentiment
Whether the tone about you is positive, neutral, or negative
80% positive framing
The core formulas are simple:
\text{Share of Model} = \frac{\text{Your brand mentions}}{\text{Total mentions of all tracked brands}} \times 100\text{Citation Frequency} = \frac{\text{Answers that cite your domain}}{\text{Total answers analyzed}} \times 100Why Each Platform Needs Its Own Approach
You cannot treat ChatGPT, Gemini, and Perplexity as one source. They retrieve information differently, show sources differently, and give different answers to the same question. Measuring them together hides which platform is helping or hurting you.
Platform
How it builds answers
How sources appear
What to watch when measuring
ChatGPT
Uses its training knowledge, and can search the web when a question needs fresh information
Links appear mainly when web search is used
Track answers with and without search, since mentions can come from memory alone
Gemini
Draws on its training and on Google Search results
Often shows source links and related results
Results can mirror Google visibility, so compare with your organic rankings
Perplexity
Searches the live web for almost every question
Numbered citations beside nearly every claim
The richest source of citation data, and the easiest to audit
Because of these differences, the same brand can score 40% on Perplexity and 10% on ChatGPT. Neither number is wrong. They simply reflect different systems, so each needs its own baseline.
Step 1: Build a Prompt Set That Reflects Real Buyers
Your measurement is only as good as the questions you ask. Start with 50 to 100 prompts that mirror how real customers research your category.
Group them by intent:
Discovery: "What are the best project management tools for small agencies?"
Comparison: "Compare Brand A and Brand B for remote teams."
Problem-led: "How do I stop missing deadlines across a distributed team?"
Branded: "Is Brand A good for freelancers?"
Local or niche: "Which agencies offer technical SEO in Denver?"
Pull wording from sales calls, support tickets, Reddit threads, and your own search query data. Real phrasing beats guesses.
Also list the brands you will track. Include your own brand, three to eight direct competitors, and common spelling variations, so the count stays accurate.
Step 2: Run Every Prompt Consistently
AI answers shift from run to run, so one test per prompt is not enough. Consistency is what turns noise into a trend.
Run each prompt multiple times. Three to five runs per prompt per platform is a practical minimum.
Use a clean session. Log out of personalized accounts or use fresh chats, so past conversations do not bias results.
Fix your settings. Keep the same model version, web search setting, location, and language for every run.
Record the date and model. Platforms update often, and a sudden score change may come from a model update, not your work.
Test on a schedule. Run the full set weekly or every two weeks, using the same prompts each time.
Step 3: Capture and Code Each Answer
Save the full text of every answer, then log the same fields each time. A simple spreadsheet works well at the start.
Field
What to record
Date and platform
When you ran it and on which tool
Prompt and intent group
The exact question and its category
Brands mentioned
Every tracked brand named in the answer
Position
Whether your brand was first, middle, or last in a list
Cited domains
Every source URL or domain shown
Your domain cited
Yes or no
Sentiment
Positive, neutral, or negative framing of your brand
Check the cited pages too. Knowing which of your pages get cited, and which third party sites get cited instead, shows you exactly where to improve.
Step 4: Calculate Your Core Metrics
Once the data is coded, the math is straightforward. Here is a worked example for one platform after 100 prompt runs.
Brand
Mentions
Share of Model
Your brand
30
25.0%
Competitor A
45
37.5%
Competitor B
25
20.8%
Competitor C
20
16.7%
Total
120
100%
Your brand has 30 of 120 total mentions, so Share of Model is 25%. If your domain was cited as a source in 18 of those 100 answers, citation frequency is 18%.
Calculate these numbers separately for each platform, then again for each intent group. Overall averages can hide a weak spot, such as strong branded visibility but near zero presence in discovery prompts.
Keeping this process running takes steady effort. Many companies bring in a remote SEO and AI search team in denver to build the prompt library, run the tests, and turn the raw data into clear monthly reports, while in-house staff stay focused on creating content.
Platform Specific Measurement Tips
Measuring on ChatGPT
Test both modes: answers with web search on and answers from the model's built in knowledge. Report them separately.
Mentions without links are common, so rely more on Share of Model than on citation counts here.
Ask follow up questions such as "Which one would you pick?" to capture recommendation rate.
Measuring on Gemini
Compare your Gemini results with your Google organic rankings for the same topics. Strong overlap suggests classic SEO is carrying your visibility.
Record whether sources shown are your pages, review sites, or forums.
Run prompts in the same location and language each time, since results can vary by region.
Measuring on Perplexity
Log every numbered citation. Perplexity gives the most complete source list, so it is ideal for citation frequency analysis.
Note which competitor pages and third party sites get cited most. These are your content gaps.
Re-run prompts after you publish or update a page to see how quickly new content gets picked up.
What Good Looks Like
There is no universal benchmark, because it depends on your category and how many competitors you track. Use a clear baseline first, then aim for steady gains. Beating the average competitor in your set, or closing the gap with the category leader, is a more useful goal than hitting a fixed percentage.
Manual Tracking vs Automated Tools
You can start with a spreadsheet and a free account on each platform. As your prompt set grows, manual testing becomes slow, so many teams move to dedicated AI visibility tools that run prompts on a schedule and calculate the metrics for them.
Approach
Best for
Trade-off
Manual spreadsheet
Pilots and small prompt sets under 50
Time consuming and harder to repeat
Scripted runs through platform APIs
Teams with developer support
Needs setup, and API answers can differ from what users see in the app
Dedicated AI visibility tools
Ongoing tracking at scale
Costs money, and methods vary, so check how each tool runs its prompts
Whichever route you choose, verify a sample of results by hand every month. This confirms that your tool or script is counting mentions and citations correctly.
Reporting and Review Cadence
Share the numbers in a simple monthly report so stakeholders can act on them.
Headline metrics: Share of Model and citation frequency per platform, with change from the last period.
Competitor view: Who gained or lost share, and on which prompts.
Source analysis: The pages and third party sites cited most often.
Actions taken: New content, updated pages, and earned mentions, so you can link effort to results.
Look at trends over several months, not single weeks. Because AI answers vary, small swings are normal and should not trigger panic.
Common Mistakes to Avoid
Testing each prompt only once. A single answer is an anecdote, not a measurement.
Only tracking branded prompts. Asking about your own brand tells you little. Discovery prompts are where you win or lose new customers.
Mixing platforms into one score. Each platform behaves differently and needs its own baseline.
Ignoring personalization. Logged in history can skew answers, so use clean sessions.
Changing the prompt set too often. Keep a stable core set so trends stay comparable, and add new prompts in a separate group.
Counting mentions but ignoring sentiment. Being named in a negative comparison is not a win.
Skipping manual checks. Automated counts can miss brand name variations or misread sources.
Conclusion
Measuring Share of Model and citation frequency comes down to a repeatable routine: build a realistic prompt set, run it consistently on ChatGPT, Gemini, and Perplexity, log every mention and source, and calculate the metrics for each platform separately.
Start small with a spreadsheet and a baseline. Then review monthly, study which sources the AI trusts in your category, and improve your content and earned coverage to match. Over time, those numbers show whether your brand is becoming a default answer in the places your buyers now ask for advice.
Comments