How to Do Share of Model: A Step-by-Step Workflow
You measure share of model the same way you’d measure share of voice, just with a chatbot instead of a search results page: pick a fixed set of prompts, run them across the AI platforms your buyers use, log who gets mentioned and how, then turn that log into a repeatable number you can track month over month. The hard part isn’t the math, it’s the discipline of running the same prompts the same way every time so the number actually means something.
Below is the workflow we run for our own clients at Salterra Digital Services, stripped down to the steps that matter. No fancy stack required, you can build the first version in a spreadsheet this afternoon.
Step 1: Define the Category and the Competitor Set
1. Write down the exact category you're being evaluated against
Before you touch a chatbot, define the “aisle” you’re standing in: not your brand name, but the category a buyer would use to solve their problem, “email marketing software for small business,” “regional accounting firms for construction companies,” “best waterproof hiking boots.” Share of model only means something relative to a category, because that’s the frame the AI uses when it decides who to mention.
2. Build your real competitor list, not your wish list
List every brand you actually lose deals to, plus the two or three brands that always show up in review roundups even if you’ve never crossed paths with them directly. Five to ten competitors is plenty. This list becomes the scoring key for every prompt you run, so revisit it quarterly since AI answers surface challengers faster than traditional search does.
Step 2: Build a Representative Prompt Set
3. Write solution-aware prompts
These are the prompts a buyer types when they know their problem but haven’t picked a solution yet: “how do I track backlinks without paying for an enterprise tool,” “what’s a good CRM for a five-person agency.” No brand names. This is where most of the buying journey happens, and it’s the prompt type most brands forget to test because it doesn’t feel like it’s “about them.”
4. Write comparison prompts
These mirror classic “best X” and “X vs Y” search queries: “best local SEO agencies for dentists,” “Ahrefs vs Semrush for keyword research.” Comparison prompts are where share of model gets most competitive, because the model is forced to pick a short list.
5. Write branded prompts
Include prompts that name your brand directly (“is [brand] any good,” “[brand] reviews”) alongside prompts naming a competitor. Branded prompts tell you what the model already believes about you; competitor-branded prompts tell you whether you show up as an alternative when someone’s already leaning toward a rival.
6. Lock the prompt set at 15 to 30 prompts
Mix roughly half solution-aware, a third comparison, and the rest branded. Write them in your customer’s language, not internal jargon, then freeze the list. You’ll add prompts over time as new use cases emerge, but the core set needs to stay stable so tracking is apples-to-apples across cycles.
Step 3: Choose the AI Platforms to Test
7. Cover the platforms your buyers actually use
At minimum, run your prompt set through ChatGPT, Google’s AI Overviews (and AI Mode where available), and Perplexity. Add Claude and Gemini’s standalone chat if your audience skews technical or enterprise. Each platform pulls from different sources, so treat them as separate channels, not one combined score.
8. Decide whether you're testing logged-in or logged-out, and note it
Personalization and memory can skew results, especially in ChatGPT and Gemini, where prior conversations quietly bias the answer. For a clean baseline, run prompts in a fresh or logged-out session where the platform allows it. Want to see what logged-in customers get? Run a second, labeled pass, don’t mix the two into one number.
Step 4: Collect the Responses
9. Run every prompt on every platform and save the raw output
Paste full responses into a tracking sheet, one row per prompt per platform per date. Don’t summarize as you go, capture the actual text and any cited sources or linked domains. You’ll need the raw output later to check your own tagging and catch competitor messaging that’s made its way into the model’s phrasing.
10. Run each prompt more than once before you conclude anything
AI outputs vary run to run, especially in chat interfaces without grounding. Run each prompt two or three times per session and treat the result as a range. A brand showing up in one of three runs is a weak signal; three of three is a real position.
Step 5: Score Mentions, Citations, and Sentiment
11. Tag every mention with a simple presence score
For each response, mark whether each brand on your competitor list appears at all, and if so, where: named in the first paragraph, buried in a longer list, or absent entirely. A brand named early and alone is worth more than one wedged into a ten-item list, so a simple 0/1/2 scale (absent, listed, featured) beats a flat yes/no.
12. Tag citations separately from mentions
Some AI answers link out to sources, others don’t cite anything visible. Where citations exist, note which domains got linked, this tells you what the model is reading to form its answer, a direct signal for what content and PR to invest in next. A brand can be mentioned without being cited, and cited without being named; track both.
13. Score sentiment on a simple three-point scale
Positive, neutral, negative is enough resolution for most brands. Watch for neutral-sounding mentions that quietly favor a competitor (“X is popular, though some find it expensive, Y is often recommended as a simpler alternative”), that nuance is where AI answers do reputational work a basic mention count will miss.
Step 6: Calculate Share of Model
14. Compute the basic formula
Share of model = number of prompts where your brand appears, divided by total prompts run, expressed as a percentage. Calculate it per platform first, then blend into an overall number if you want a single headline figure. Appearing in 9 of 24 ChatGPT prompts is a 38% share on that platform, illustrative math, not a benchmark to chase.
15. Weight the number if you want more signal
A raw appearance count treats a buried mention the same as a featured one, which understates real dominance or weakness. Multiply your 0/1/2 presence score by prompt count instead of using a flat percentage, and you get a weighted share that better reflects how prominently you’re being recommended.
16. Calculate it for your top two or three competitors too
Your share of model only means something next to theirs. A 40% score against a category leader at 85% is a real gap to close; that same 40% against a field where everyone else sits under 25% means the priority shifts to defense, not growth.
Step 7: Set a Tracking Cadence
17. Re-run the full prompt set on a fixed schedule
Monthly is the right rhythm for most brands, AI answers shift faster than organic rankings but not so fast that weekly tracking adds signal over noise. High-velocity categories can justify biweekly checks; slower B2B categories can often get away with quarterly.
18. Keep a running log, not just a snapshot
Store each cycle’s results in the same sheet rather than starting fresh each time. The trend line, is presence climbing after a content push, did it drop after a competitor launch, is worth more than any single month’s number, and it’s the only way to prove this work is moving anything.
Step 8: Close the Gaps
19. Fix absence before you fix ranking
If you’re missing entirely from a solution-aware prompt, that’s a content gap, not a prominence problem. Publish a page or article that directly answers the underlying question in plain, structured language, the kind of content this whole site is built around, before worrying about how you’re worded once you appear.
20. Fix framing where you're present but underweighted
If you show up but always third or fourth, always paired with the same qualifier (“a solid budget option”), that language is coming from somewhere the model trusts, review sites, comparison articles, forum threads. Find that source and improve your standing there, or publish a stronger, more citable alternative the model can pull from instead.
21. Fix negative or stale sentiment at the source, not in the prompt
You can’t argue with a chatbot, so don’t try to prompt-engineer your way to better sentiment. Trace the negative framing back to the review, forum thread, or outdated page it’s likely drawing from, and address it there: respond to the review, update the stat, publish current proof. The model updates when its sources do.
Frequently Asked Questions
How many prompts do I need for a reliable share of model number?
Fifteen to thirty prompts, run two or three times each per platform, is enough for a directional read for most small and mid-sized brands. Larger brands tracking multiple product lines or regions may need separate prompt sets per segment.
Can I automate this instead of doing it manually?
Parts of it, yes. Several rank-tracking and brand-monitoring tools now offer AI visibility modules that automate the prompt-running and mention-logging steps. We still recommend a manual pass at least once a quarter, automated tools can miss nuance in sentiment and framing that a human reader catches instantly.
Do I need to track every AI platform, or can I focus on just one?
Start with whichever platform your buyers are most likely to use for research, for most B2B and local service categories that's ChatGPT and Google's AI features. Expand once you have a working process, tracking one platform well beats tracking five badly.
What counts as a "good" share of model score?
There's no universal benchmark, it depends entirely on your category and how many real competitors are in it. The number that matters is your trend relative to your own past scores and to your top competitors, not a fixed percentage to hit.
How is this different from tracking traditional search rankings?
Traditional rank tracking measures position on a results page for a fixed query. Share of model measures whether and how you're described inside a generated answer, which can vary between identical prompts and pulls from a different, often narrower, set of sources. Treat them as complementary metrics, not substitutes.
What do I do if a competitor is clearly winning the comparison prompts?
Read exactly how the model frames the win, then check what content and third-party sources are likely feeding that framing. Usually it traces back to stronger comparison content or clearer category positioning on their end. Close that specific gap first rather than trying to out-optimize the model in the abstract.
Terry has 30+ years in software and SEO. He’s the founder of Salterra Digital Services and SEO Spring Training, host of the Roundtable SEO Mastermind, and lead instructor at SEO University — teaching the exact tactics his team uses on client work.
Continue learning
Ready to master this?
This guide is one lesson from the Measuring AI Visibility Share of Model course. Get every lesson, framework and checklist — plus the full 38-course catalog — inside SEO University.
Practitioner-focused training across the full digital marketing stack — from technical SEO to conversion optimization and the AI search era. By Salterra Digital Services, since 2011.