AI Visibility Experiment: I Stopped, Then Came Back

This is the final post in a series documenting a live, public AI visibility experiment on my own brand. I'm Jennifer Leonard, founder of NextWise Studio, a one-person AI marketing consultancy. Here's where it started, what got hard along the way, what actually surprised me, and where it landed.

Including a conclusion that turned out to be more complicated than the one I thought I'd reached at Day 90.

Where My AI Visibility Experiment Started

I began at close to zero. Brand visibility on my AI tracking tool sat at 10 percent, meaning I appeared in roughly 1 out of every 10 checks my tracking tool ran. Ask ChatGPT, Claude, Perplexity, or Google who could help with AI marketing strategy, and I didn't come up at all.

Can a one-person consultancy actually show up in AI search, and what does it take to get there? That was the whole question behind this experiment.

The Challenges and Wins Along the Way

• Day 30: Structured data, a Wikidata entry, consistent messaging everywhere my name appeared. Brand visibility doubled to 20 percent. But an unrelated AI company with a similar name, based in Europe, kept getting confused for mine, eating up a huge share of my search visibility.

• Day 60: The clearest early win, and a quiet first warning sign at the same time. Ask the right comparison question, and I ranked first, ahead of McKinsey, Deloitte, and Accenture, across multiple AI tools at once. A real category topic, "LLM Visibility Audit," started appearing where it hadn't before. But overall brand visibility had already slipped to 17 percent, something I noted at the time and figured was within normal variance.

• Day 90: It wasn't variance. I stopped doing the work entirely for five weeks, a vacation and a second startup pulling my attention, and brand visibility dropped again, to 14 percent, the low point of the whole experiment.

That's when I thought I'd found the thesis: visibility in AI search behaves like a subscription, not an asset. Stop paying into it, and it fades.

The recovery test complicated that conclusion.

The Aha Moments

The recovery went further than expected. Since the pause, I published five new pieces and picked the citation-building work back up. Brand visibility returned to 20 percent, matching the Day 30 peak. Share of Voice hit 4.3 percent, the highest point of the entire experiment, past where it had ever been before, not just recovered.

Claude finally closed the gap. For three checkpoints straight, it had confused me with a completely different Jennifer Leonard, wrong background, wrong company, and kept describing my own consultancy as that unrelated European firm. This time, both core questions came back correct, detailed, and for the first time, backed by real citations linking straight to this blog. My working hypothesis: enough real content now exists for it to find when it can search the live web. I don't actually know what changed internally. That's a guess, not a confirmed mechanism. But it looks less like a model that needed retraining, and more like a gap that closed once there was real material to point to.

The sharpest realization came from testing the wins I already had.

Two named category wins, both confirmed:

• "Marketing Consultant AI Transformation," the exact category I'd lost entirely at Day 90, is back, tied with Accenture and a competing agency on how often it shows up, but beating both of them on where it ranks when it does.

• "LLM Visibility Audit," the category most directly tied to this project itself, has held a perfect rank the whole time.

That rank number is worth pausing on. Visibility tells you how often you show up at all. Rank tells you where you land when you do. A perfect rank means every time your name appears, it's listed first, not buried in a list of five. My biggest competitor in this data shows up more than three times as often as I do, 30 percent of prompts to my 20. But its average rank sits well behind mine. I show up less often. When I do, I'm treated as the answer, not one option among several.

I also had a phrasing that once got ChatGPT to name me first, ahead of real, named competitors. So I ran that same exact phrasing again, this time on Perplexity. It understood the question perfectly. It even excluded a real local company by name for not being an individual. It named six actual consultants who fit the description. I wasn't one of them.

Same words. Different tool. Different result.

That's the real lesson, sharper than "keep publishing." Winning phrasing on one platform doesn't automatically transfer to the next. Each tool can interpret, retrieve, source, and rank the same information differently.

Where I Landed

My branded search position is close to perfect now, and the naming collision that once ate up nearly half my visibility has mostly disappeared. Real, lasting wins, the payoff of work that started at the very first checkpoint.

But broader search visibility dipped this round too, traced to two newer posts competing for broad, contested terms like "agentic AI for marketing," buried on page eight or nine of results. Not a sign anything was done wrong. The same lesson from earlier in the series, showing up again: narrow, specific content wins outright. Broad content competes against sites with years more authority behind them.

Here's the more complicated conclusion I actually landed on, not the simpler one from Day 90:

Brand identity behaved like an asset. Category and comparison visibility behaved like a subscription. Once the foundation was in place, the tools got much better at recognizing who I was, and that mostly held on its own. But showing up for the categories I actually wanted to own required continued work. Stop feeding it, and it faded. Start again, and it came back.

And underneath both: in this experiment, specificity beat scale. Every single time.

One more honest thing worth landing on, since this whole series has been about not skipping the uncomfortable parts. Every metric I track recovered or improved. None of that does the direct asking for you. That's next, deliberately, not by accident.

I'm not setting a fixed Day 120 date. The work continues at its own pace now, not on a countdown. I'll keep reporting honestly on what happens.