AI Visibility · Case Study
There is a particular kind of credibility risk in telling businesses to measure their AI visibility without having measured your own. So before offering this to anyone, we ran it on ourselves. Here is what came back, including the parts that were not flattering.
By Izzy Gregorio · Updated August 2026 · 8 min read
In short
We tested our own agency across ChatGPT, Perplexity, and Gemini using the queries a prospect would actually type. The result was inconsistent rather than absent: present in some phrasings, missing in others, with competitors appearing more reliably despite thinner positioning and less content. The cause was external presence rather than content quality. Generative Engine Optimization is mostly about your footprint outside your own website.
The test
The same test we now run for clients. A series of queries across ChatGPT, Perplexity, and Gemini, built around how a prospective client would genuinely search rather than around the keywords we would prefer to rank for. Things like marketing agencies for faith-based organizations in Southern California, or AI marketing strategy for small businesses in Temecula, plus several variations on each.
Every result documented. Four things recorded for each one.
| What we recorded | Why it matters more than a ranking would |
|---|---|
| Did we appear at all | There is no page two here. You are either in the answer or you do not exist in that conversation, and nobody scrolls to find out otherwise. |
| In what context | Named as a recommendation is different from appearing in a list of options, which is different again from being mentioned in passing. |
| How favorably described | The description a model gives is often your first impression now, and it was assembled from sources you did not choose. |
| Who appeared instead | The most useful column. Knowing who takes your place, and being able to work out why, is where the fix list comes from. |
Audit conducted [MONTH YEAR]. AI model outputs change frequently, so results represent a specific point in time rather than a fixed position.
The result
Going in, I would have guessed we would do reasonably well. We are a relatively new agency, but we have been deliberate about positioning, we publish consistently, and we are active professionally. I expected moderate visibility with some gaps.
What came back sat closer to the lower end of the range we now tell clients is typical for a first audit. A real presence, and an unreliable one. In some query variations we appeared, sometimes described well, sometimes just listed. In others we did not appear at all, while competitors with what I would consider comparable or less developed offerings did.
The inconsistency was the finding. Not invisible, not reliably visible. It depended on which tool, which exact phrasing, and honestly which day. That turns out to be an extremely common result, and it points at a specific problem rather than a general one.
It is an entity clarity gap. Enough presence to register sometimes, not enough consistency to register dependably. Which is a very different problem from having no presence, and it has a different fix.
The uncomfortable part
Here is the genuinely useful part, and it was not comfortable. When we ran the same queries against a few other agencies in adjacent markets, several appeared more consistently than we did, despite what looked to me like less specific positioning and considerably less published content.
When we looked into why, one pattern explained most of it, and it had nothing to do with the quality of what anyone had written.
They appeared in local business directories. Unglamorous, easily dismissed, and apparently load-bearing.
Their name, address, and phone matched everywhere. Identical strings across every platform, which is the single cheapest thing on this list and the most commonly skipped.
They had been referenced by other people. Local press, community contexts, partner sites. Someone other than them had said their name in public.
That is not a criticism of those agencies. If anything it is a useful correction, because it confirmed something I had been saying somewhat theoretically and had not fully absorbed myself.
Generative Engine Optimization is not primarily about your content. It is about your presence in the broader conversation.
And that presence can lag well behind strong content and clear positioning, if nobody deliberately built it. You can publish for two years and still be a stranger to a model, because a model is reading what the world says about you, not only what you say about yourself.
The fix list
None of it was dramatic. It is the same sequence we now recommend to clients, and the order matters more than the individual items, because each stage makes the next one work.
Name, description, and category made identical across our website, Google Business Profile, professional profiles, and every directory we appear in. Not similar. Identical, character for character. Variation reads to a machine as uncertainty about who you even are, and uncertainty loses.
FAQ schema on the pages that matter, addressing the questions a prospective client actually asks, with answers written to make sense lifted out of context. This is the cheapest way to give a model something clean to quote.
Legitimate third-party mentions. Guest content, local business associations, partner cross-references. Real relationships and real contributions rather than anything purchased, because the shortcuts here are visible and they age badly. This is the slowest stage and the one that actually moved the competitors ahead of us.
The sequence in one line: foundation, then citation-worthy content, then distribution and external presence. Reversing it is the common mistake. Chasing mentions before your own information is consistent means every new mention adds another slightly different version of you to the pile.
Keep going
What we are testing, what came back, and what we changed because of it. Written the way this post was written, which is to say with the unflattering parts left in.
No spam. Unsubscribe anytime.
Why publish this
Publishing a mediocre result about your own agency is an odd marketing decision. Here is the reasoning.
We tried it on ourselves and here is what happened is more useful than advice offered from a distance, even when the outcome includes we were not as visible as I assumed. Especially then, actually.
The middling result is the normal one. Most businesses we have audited since land in a similar place: some presence, inconsistent, with specific identifiable gaps. If you are worried an audit will surface something embarrassing, the far likelier outcome is a mixed picture with clear fixes. That is information, not a verdict.
Willingness to look at your own data is what this shift asks of you. The temptation is to assume you are probably fine, because the search work is fine, the content is fine, the brand is fine. Those are necessary and they are no longer sufficient, and the only way to know is to look.
We did not love everything we found about ourselves. We would rather know than not, and I suspect most owners feel the same way once they have seen their own results in front of them.
Common questions
A structured test of how AI tools describe and recommend your business. You run the queries a prospect would actually type across several assistants, then record whether you appear, in what context, how favorably you are described, and who appears instead of you. The competitor column usually produces the most actionable findings.
That is an entity clarity gap, and it is the most common finding in a first audit. You have enough presence to register sometimes but not enough consistency to register reliably, so results vary by tool, by phrasing, and by day. The usual cause is inconsistent business information across platforms rather than weak content.
Usually because their footprint outside their own website is stronger. Directory listings, consistent name and address and phone details across platforms, and references from local press, partners, or community organizations all build entity signals. AI answers draw on what others say about a business, not only what the business publishes itself.
Consistency, then structure, then external presence, in that order. Make your name, description, and category identical everywhere. Add structured answers to your core pages. Then pursue legitimate third-party mentions. Reversing the order wastes the mentions, because each new one adds another inconsistent version of you.
Quarterly is a reasonable rhythm for most businesses. AI model outputs change frequently, so any single audit is a snapshot of one specific period rather than a fixed position. Re-testing the same queries over time also shows whether the fixes worked, which a single test never can.
Start here
Conspicuouz Creative Group (CZ Creative Group) tests your brand across ChatGPT, Perplexity, and Gemini, compares you against your closest competitors, and returns a prioritized list of specific gaps. The same process we ran on ourselves, in the same format, with the same honesty about what it turns up. Free, and yours to keep.
Not ready yet? Subscribe to the newsletter and get the next breakdown in your inbox.