What we asked
Two kinds of prompts. Discovery prompts are the questions a future client asks before they know us: a branding studio, an agency for a startup identity. Branded prompts contain our name. Each prompt went to ChatGPT, Perplexity and Gemini, three times each, with web search on.
Eight discovery prompts × three engines × three runs gives 72 answers. We kept every one of them.
What came back
Across all 72 discovery answers, Brandson was not named once. Not by one engine, not in one run. The 13 branded answers were the opposite: every engine recognized us and described the studio correctly.
Being recognized by name helps only people who already know us. New clients come from questions that don’t include our name.
Who the engines trusted instead
When an engine searches the web before answering, it cites its sources. For our discovery prompts, three sites came up again and again:
| Cited site | Answers citing it |
|---|---|
| atico3.com | 11 |
| clutch.co | 9 |
| sortlist.com | 9 |
These are the pages the engines read before recommending a studio, so a studio that wants to be recommended needs to be listed there.
The name problem
Brandson is not a unique name. brandson.com belongs to someone else, and a Dutch company uses the same name. When the signals that tie a name to one company are weak, an engine can’t be sure which Brandson you mean, so on discovery questions it plays safe and names someone else.
Our own site linked only one profile in its Organization.sameAs markup: LinkedIn. That is a thin trail to follow.
What we are changing
- Get listed, accurately, on the sites the engines cited.
- Link every real profile from sameAs and keep the name, address and description the same everywhere.
- Write pages that answer the discovery questions directly, in plain words.
- Keep the same prompts running daily, and come back here with the next numbers, whatever they are.
We won’t claim any change caused a result. We will show when the numbers moved and what we did around that time.
