What one AI answer can and can’t tell us about brand visibility

A playful question, one surprising answer, and the careful reading it deserves.

In May 2026 we asked an AI search tool a deliberately silly question about our own name. The answer was surprising, and it made a good story. It also deserves a more careful reading than we gave it at the time.

The observation

The prompt was a casual, half-joking question:

“what’s the AI agency that sounds like bacon?”

The tool’s reply, as reported in the original post, was along the lines of “You’re probably thinking of BayconAI — the San Francisco AI agency whose name sounds like ‘bacon.’” It cited builtinsf.com, a third-party directory, rather than bayconai.com.

Because those details weren’t kept, you can’t reproduce this observation from what we published. Treat it as an anecdote, not as a verified result.

What it does and doesn’t measure

Three distinctions matter here.

• Branded recall versus unbranded discovery. The prompt described a name (“sounds like bacon”). That tests whether a tool can connect a description to a name it has seen. It says nothing about whether someone asking an ordinary question, such as who can help with AI at their company, would be pointed to BayconAI.
• Citation versus recommendation. Naming a business, citing a page about it, and recommending it are different things.
• Either versus commercial impact. Even a clear recommendation doesn’t tell you whether anyone got in touch or became a customer.

Attribution accuracy

The answer cited a third-party directory instead of our own website, and described us as a San Francisco agency. The original post stated that we are not based in San Francisco, so that detail was inaccurate. We don’t know why the tool did either. It may have relied on the directory because that page was easier for it to find, it may have generalized from the Bay Area, or it may simply have varied from one run to the next.

Alternative explanations

Before crediting any particular cause, consider the ordinary ones:

• AI answers differ between runs, between tools, between modes with and without web search, and between locations and dates.
• A directory listing may simply have been published earlier or been easier to read than our own pages.
• A playful, narrow prompt is easy to match to a distinctive name.
• Nothing we did to our own website can be separated from any of the above based on one answer.

How to test AI visibility more carefully

1. Write down a fixed set of realistic prompts in advance, including the ordinary questions a customer would ask.
2. Keep branded prompts (that name your business) separate from unbranded ones (that don’t).
3. For every run, record the tool and search mode, the date and time, the location context, the sources cited and whether the answer was accurate.
4. Repeat the same prompts later and compare like with like.
5. Report what changed without assuming each change was caused by your own edits.

What is useful to improve

None of this guarantees a mention, but these are sensible, low-regret steps:

• Say clearly on your own website who you are, where you work and what you do, in visible text.
• Add structured data only where it matches what the page visibly says.
• Keep your business details consistent across the profiles and directories you control, and ask for corrections where other sources are wrong.

Limitations

This is one observation, from the business being discussed, that was not independently recorded. It shouldn’t be used to draw conclusions about how AI tools choose which businesses to mention. Those choices vary and aren’t fully visible from the outside.