The Houston Ad Voice

A Company Tried to Sell Me a $10,000 AI Marketing Tool, So I Built the Tool Instead

What one sales pitch taught me about AEO, GEO, and matching your marketing tools to what your business actually needs.

AI marketing tools are everywhere right now, and there’s no clear map for using them well. Long videos from self-styled AI gurus promise a shortcut to mastering all of it, while developers who already know these tools sell that same expertise back to marketers who don’t.

The shift that caught my attention is closer to home. Customers now reach a recommendation two ways that used to be separate. Some ask a conversational assistant like ChatGPT out loud or by typing. Others just run a normal search, where an AI-generated answer now sits above the links and does the same summarizing work before anyone clicks through. That holds whether the audience is consumer-facing, business-facing, or government-facing. When someone asks an AI assistant to recommend a service like theirs, does the brand show up, and is it described correctly?

Every marketer in Houston I know is being pitched the same promise: urgency, automation, dashboards, future-proofing, scaled to whatever budget a shop runs on, from an enterprise Marketing Cloud instance. That raises a real question for anyone not running a large team. Does the budget exist to buy every tool the industry says is required?

That’s the backdrop for the pitch that got me thinking about all this. A company tried to sell me on moving my marketing operations to a single platform for more than $10,000 a year, built for scalable content generation, brand monitoring, web development, and built-in AEO and GEO measurement. I’ll leave the platform unnamed. The assumption behind that price is what stuck with me. AI answer engines are already shaping how brands get found, summarized, compared, and recommended.

Rather than accept or reject the pitch outright, I treated it as something to test and learn from: could a marketer with limited time and a limited budget learn what this platform promises to deliver by building a rough version of it first?

Traditional search trained marketers chase rankings and keywords. AI answer engines work differently, instead of handing back a page of links, they read the web directly and compress a dozen sources into the one answer they give the person who asked, so the brand has to make sense to a person and to the model summarizing it on that person’s behalf.

This is the same shift already reshaping developer tooling for a different audience. Coding assistants like Claude Code and Cursor read a company’s documentation the way an AI answer engine reads marketing content, fetching pages directly and either using them or skipping past them in seconds. Marketing AEO and GEO ask a similar question: can an AI system find what a business has published and trust it enough to repeat it, without a person there to explain the context?

The $10,000 pitch was selling a feature for tracking those exact questions at scale.

So I ran a AEO/GEO test on something with zero connection to my client as a play ground to see how the search results and data stacks up: who is Houston’s most iconic mascot?

In a market this crowded, a name alone doesn’t cut it. A firm with a consistent character, name, and visual identity across its site, case studies, and press gives an AI model something repeatable to point to. A thin public record gives it nothing to cite, no matter how good the work is.

All three engines landed on the same final four, Orbit, Clutch, TORO, and Shasta, then flat-out disagreed on who deserves the crown.

ChatGPT crowns Orbit, “the lime-green space alien… a staple at Astros games.”

Gemini puts Clutch first and Orbit second, calling him “deeply embedded in Clutch City basketball culture.”

Perplexity also backs Orbit, calling him “especially prominent” over Clutch and TORO.

Orbit takes two out of three, and each engine states its pick with total confidence and real sources behind it. The final four is settled, decades of press and Hall of Fame entries make sure of that, but the ranking on top isn’t, so every engine fills that gap from whatever it read that morning and states it like fact. Gemini even volunteers, flatly and unprompted, that Houston is the only U.S. city with three mascots in the Mascot Hall of Fame.

The same mechanic applies to any client’s category. In a surplus market with dozens of firms saying the same things, making the list at all comes from having a public record thick enough to repeat. Past that point, the answer is whatever the model assembled that day, and the customer only ever sees the one version it gives them.

Houston’s Most Iconic Mascot, depending which AI you ask. One run of one question is an illustration, not a study, and the same instability this piece describes applies to the test itself.

Going back to the pitch I sat through, I was promised scalable content generation, brand monitoring, direct AEO and GEO visibility measurement, and a clearer line between what a business publishes and what an AI system says about it in return. That premise sent me down this build, and I wanted to know whether those promises of these features held up before paying for them.

I ended up building a script that runs a list of questions against to ChatGPT, Perplexity, and Gemini every month and syncs every result straight into my client’s workspace to track how they preform. That’s the premise of what these platforms offer, AEO and GEO tracking, and for more than the $20 a month that feature costs inside most of these third party platforms.

Built for my personal use, my script pairs to my clients Notion-based prompt library with a results log, test questions run across ChatGPT, Perplexity, and Gemini every month, and a Claude-based pass then classifies each answer and tags competitors. A free GitHub Actions job holds the schedule and can be triggered manually on Notion using a button for an ad hoc check which can run the check anywhere (mobile or desktop); worst-case scenario my API spend across all three providers caps around $10 a month, though after running this several times the actual usage has stayed under $1. The real value came in once this ran against my architecture-firm client for a few weeks. Across 86 logged answers, the firm showed up by name in 27 of them, a 31 percent visibility rate that shifts by engine: Gemini names the firm 42 percent of the time, Perplexity and ChatGPT closer to 27 percent. When the firm does get named, the sentiment is good: 25 positive mentions, zero negative.

The competitor log makes the size of that gap concrete. One ChatGPT answer alone listed 14 architecture firms for a single specialized-facility question and named my client in none of them. Across every logged run, the most-mentioned competitor turns up 28 times, more than double my client’s own count, with a long tail of other firms filling the same slots my client doesn’t. The same mechanic drives the mascot test: in a category this crowded, mention rate simply tracks the size of a firm’s public record, and a competitor with a decade of published projects has more surface area to get repeated. Every run gets logged with a sentiment tag, the competitors named, and the sources cited, the same fields a paid share-of-voice dashboard tracks, just without the subscription.

That gap points to a content strategy. Publish more of the specific project types AI systems keep citing competitors for, starting with the facility type that came up empty in this test. The log becomes a topic calendar, since whichever project type keeps surfacing competitors and not my client is the next case study, blog post, and social recap worth pitching. Written and shared consistently, that kind of public record is what these engines already reward, and it does double duty for a client’s search ranking.

The right choice comes down to scale. A full platform makes more sense once you need shared dashboards and approvals across a bigger team, leadership expects polished recurring reports, and $10,000 a year barely registers against the rest of the budget. Building it yourself makes more sense when you’re a team of one who can read results directly, $1 to $5 a month covers what you need to know, and you’d rather learn the data firsthand than manage a new platform. Either way, it helps to price out what a single insight actually costs you to get on your own before signing up for anything larger.

Building instead of buying answered a direct question. What would watching my own client’s AI visibility teach me that a $10,000 subscription wouldn’t? Real competitors filled the search answers my client was absent from, and the log showed exactly which content gap to start fixing and producing more content. Any marketer with a laptop and some patience can build the same early-warning system for a client, ahead of the next platform pitch that assumes they need one.

For other AAF members sizing up a similar pitch, the tool isn’t the test. Ask an AI assistant what it already says about your client or your brand, see who shows up instead, and use that gap to decide what you need to buy, build, or fix first.

Every gap you find is a brief. Build the tool before you buy the platform.