Share of Model Case Study: A Step-by-Step Walkthrough
The scenario below is illustrative — a composite built from the kind of engagement we run at Salterra, not a verified client case study. We’re using it to show the mechanics of a Share of Model project end to end: how you baseline it, what “gaps” actually look like when you dig into the data, and how a team decides what to fix first. Treat every number here as a hypothetical placeholder for teaching purposes, not a claim of results.
Share of Model — the frequency and prominence with which a brand appears across AI answer engines for its category — is a newer metric, and most teams have never watched one get built from a blank spreadsheet. This walkthrough fills that gap with a step-by-step process you can adapt to your own business.
The business and the problem
Picture a mid-sized regional HVAC company — call it the client. They rank well in traditional Google organic for “AC repair [city]” and similar terms, and their local pack presence is strong. But their owner noticed something: when he asked ChatGPT and Gemini “who’s the best HVAC company in [city],” his own brand didn’t show up. Competitors he considered smaller did.
That’s the trigger for most Share of Model work. It’s rarely “we want to measure AI visibility” in the abstract — it’s a specific, uncomfortable moment where someone sees a competitor recommended by an AI assistant in a spot the client expected to occupy.
Establishing the baseline
Before diagnosing anything, you need a number to move. The baseline step looked like this:
- Built a query set of roughly 25 prompts a real customer would plausibly type into an AI assistant — service-based (“who repairs mini-split systems in [city]”), comparison-based (“best HVAC companies in [city]”), and problem-based (“AC not cooling, who should I call”).
- Ran each prompt across the major assistants the client’s customers actually use, multiple times each, to account for response variance.
- Logged three things per response: was the brand mentioned at all, where in the answer it appeared, and what was said about it when it did appear.
- Did the same for the three competitors the client named as their real-world rivals.
In this illustrative pass, the client’s brand appeared in roughly 15% of relevant prompts, almost always in a list rather than as the lead recommendation. One competitor appeared in a hypothetical 60%+ of prompts and was frequently named first. That gap — not the raw percentage — was the actionable finding.
Diagnosing the gaps
A low Share of Model number tells you almost nothing on its own. The diagnosis is where the real work happens, and it splits into a few categories worth checking every time:
Source coverage
We looked at what sources the AI systems appeared to be drawing from when competitors got mentioned — review platforms, local business directories, “best of” roundup articles, and the competitor’s own site content. In the illustrative case, the leading competitor had been featured in two locally-published “best HVAC companies” roundups and had a materially higher review count on the platforms these tools seem to weight.
Content structure
The client’s website had service pages, but they were thin — a paragraph of copy per service with no specifics about process, pricing ranges, service area detail, or the kind of concrete operational detail that gives an AI model something citable. Compare that to entity-based content: pages structured so a single passage directly answers a single question, with clear entities (service names, service areas, credentials) stated plainly rather than implied.
Consistency of entity signals
The business’s name, service list, and service area were phrased slightly differently across its website, Google Business Profile, and the handful of directories it was listed in. Individually minor; in aggregate, this kind of inconsistency makes it harder for a model to confidently resolve “this business” as a stable, well-defined entity.
Absence from comparison contexts
None of the third-party content that existed about the client positioned it against competitors. Nearly every appearance of the leading competitor, by contrast, showed up in exactly that comparative context — “top HVAC companies,” “X vs Y,” “best-rated in [city].” That’s the content format AI answer engines lean on heavily for comparison-style prompts.
Building the plan
With the diagnosis in hand, the plan prioritized fixes by leverage — what would most plausibly move the needle relative to effort — rather than trying to do everything at once. The priorities, in order:
- Rebuild the core service pages around clear, directly-answerable structure: one primary question per section, specific service details stated in plain language, and consistent naming of the business and its service area throughout.
- Standardize NAP and service-list language across the website, Google Business Profile, and every existing directory listing.
- Actively pursue inclusion in local “best of” and comparison content — pitching relevant local publications and industry directories rather than waiting to be discovered.
- Encourage a steady, ongoing flow of detailed customer reviews rather than a one-time push, since recency and volume both appear to matter to how these systems weight review-based sources.
- Publish a small set of genuinely useful, experience-based content — real troubleshooting guides, seasonal maintenance advice, and explanations of what things actually cost and why — written by people who do the work, not generic filler.
Execution
Step one: entity cleanup
Every mention of the business online got audited and reconciled to a single canonical name, phone number, address, and service list. This is unglamorous work, but it’s foundational — you can’t build citable authority on top of inconsistent facts about who you are.
Step two: service page rebuild
Each core service page was rewritten so that a reader (or a model summarizing the page) could extract a direct, specific answer to “does this company do X, and what does that involve” without inference. Vague marketing language was replaced with concrete specifics: equipment brands serviced, typical response times, what’s included in a diagnostic visit.
Step three: earned placement
The team identified the handful of local publications and industry directories most likely to be cited by AI systems for “best HVAC” style queries, and pitched the client for inclusion with a fact sheet that made it easy for an editor to write about them accurately.
Step four: review cadence
Rather than a one-off review campaign, technicians began asking satisfied customers for reviews as a routine part of every completed job, with occasional prompts encouraging specific detail about the problem and the fix — the kind of detail that reads as genuine, first-hand experience.
Step five: original content
A handful of pages were published that reflected the owner’s own operational judgment — the kind of thing only someone who runs the business would know, like how to tell if a system needs replacing versus repairing, or what a fair regional price range actually looks like.
Re-measuring
Roughly a quarter after the changes went live, the same query set was run again, using the same methodology as the baseline — same prompts, same assistants, same multiple-run approach to smooth out variance. In this illustrative scenario, brand mentions rose from the hypothetical 15% baseline into the 35-40% range, with the business beginning to appear as a named option (though still rarely the sole first-listed recommendation) in comparison-style prompts. The nature of the mentions also shifted — early responses tended to just list the name; later responses more often included a specific detail, like a mentioned specialty or service area, echoing language from the rebuilt service pages.
Worth being blunt here: this kind of movement is not linear, not guaranteed, and not something you can promise a client on a timeline. AI answer engines update their retrieval and training inputs on their own schedule, and a single re-measurement is a snapshot, not proof of a durable trend. That’s why ongoing tracking matters more than any single before/after comparison.
Qualitative outcomes
Beyond the tracked prompts, the client reported a few things worth noting even though they’re anecdotal by nature: new customers occasionally mentioned finding them through a “best of” list that hadn’t existed before the campaign, and the owner said he’d personally re-run his original ChatGPT query enough times to notice the brand showing up more consistently, if not universally. Neither of those is a hard metric, but they’re the kind of on-the-ground signal that tends to show up before the aggregate numbers fully catch up.
Lessons
A few things this kind of project reliably teaches, regardless of industry:
- Share of Model doesn’t move because of one tactic — it moves because entity consistency, citable content, and third-party validation improve together.
- The gap between you and a competitor is more informative than your raw percentage. A brand at 40% mentions with a competitor at 45% is in a different position than one at 40% with a competitor at 90%.
- Comparison-context content — “best of” lists, versus-style articles, roundups — carries outsized weight for the prompts that actually drive buying decisions.
- This is a measurement discipline, not a one-time audit. The tools and models change; a Share of Model baseline from six months ago tells you where you started, not where you stand today.
Frequently Asked Questions
Is this a real client result?
No. This walkthrough is an illustrative, composite scenario built to teach the Share of Model methodology. All figures are hypothetical and should not be read as verified outcomes from any specific engagement.
How long does a project like this typically take to show measurable movement?
It varies widely by industry, existing entity strength, and how much content and citation work is needed. Meaningful movement is more often measured in months than weeks, since AI systems don't refresh their view of a brand instantly.
Do I need a large query set to get a useful baseline?
A focused set of 15-30 prompts that reflect real customer intent is usually more useful than a huge, unfocused list. Quality and relevance of the prompts matter more than sheer volume.
Which AI assistants should I include when measuring Share of Model?
Prioritize the ones your actual customers are likely to use. For most local and service businesses, that means covering at least the major consumer-facing assistants, since coverage and behavior differ meaningfully between them.
Can I run a Share of Model audit myself without an agency?
Yes, in principle — the methodology is mostly disciplined process rather than proprietary tooling. What takes practice is diagnosing why the gaps exist and prioritizing fixes correctly, which is where experience shortens the timeline.
Does improving Share of Model hurt traditional SEO performance?
No. The underlying work — entity consistency, clearer content structure, genuine reviews, and third-party validation — tends to support traditional search performance as well, since both systems reward clarity and credibility.
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.