A word on why I am writing on an agency’s blog.

The Visibility team runs its client work on Brault: one library per client, with the campaign files, the brand assets and the approved versions in the same place.

When a banner or a video is ready, the account lead shares a link that carries Visibility’s own logo and colours, the client leaves a comment on the exact frame or pixel, and the approved version is the one that goes to the printer or the ad account.

Big deliveries go out through the same library as a transfer, up to 250 GB, from a link that does not expire in a week. A board next to the files shows which pieces are in review, approved or delivered, so nobody asks in a chat.

The part I like most is what they built on top: through Brault’s public API and its n8n node, a client approval in Brault fires a small workflow that pushes the approved file to the client’s Meta or Google Ads asset library, posts a note in that client’s Slack channel and moves the card on the board to “delivered”.

When a new client signs, another workflow creates the library, the board and the access roles from their CRM record. None of it needed a developer; the n8n node and the API docs were enough.

Before that, the same work lived across a shared drive, a transfer service, a review tool and a spreadsheet. I know that setup well: before Brault, my co-founder Pablo and I ran creative and advertising agencies, and most of what the product does is what we wished we had had on a Friday afternoon with a client waiting. We got talking about a different problem, whether an AI assistant would ever recommend a small company like ours, and this article is the answer I owed them.

I co-founded Brault, a creative file management platform: AI search, libraries, review and approval, built in file transfer and white label branding.

In September we started asking ChatGPT, every Monday, the questions our buyers ask it. Here is what changed our minds, and what I would tell a marketing team that wants an assistant to recommend it and not only Google to rank it.

The habit: A fixed set of questions, asked the same way each week

We wrote down a dozen questions a buyer in our category might type into an assistant: alternatives to a named competitor, a tool for a specific job, and a few that describe what we offer without naming us. Every Monday we ask them to ChatGPT with web search on, from a clean session, and we log three things per question:

  • Are we cited, meaning one of our pages shows up in the sources?
  • Are we recommended, meaning the answer names us as an option?
  • And which websites was the answer built from?

That third column is the one most teams skip, and it is the one that explained almost everything. Within a few weeks, several questions had changed state in both directions, and each change had a visible cause in that column.

1. The assistant reads your page in the buyer’s words, not yours

The first question that flipped in our favour was one that described our product exactly, in the buyer’s language. For weeks the answer named a competitor whose page title used those same words. We had the same features on a page written in our own product vocabulary, and it was never retrieved.

We rewrote that page: same URL, new title and opening paragraph in the buyer’s phrasing. The following week the assistant quoted the new title back at us and named Brault as the closest fit. Nothing else on the page changed.

If you do one thing for a client this month, run their value proposition through the questions their buyer would type. When the words do not match, the assistant has no reason to pick that page over one that does.

2. Lists get picked up. Head to head pages do not

We publish two kinds of comparison content: “competitor versus Brault” pages, and “best alternatives to a competitor” roundups where Brault is one entry among several.The roundups were retrieved within days for the “alternatives” questions.

The head to head pages were not retrieved for those questions at all, only for queries that already contained our brand name. An assistant that is building a list wants a source that is already a list.

3. You can lose a mention to a directory that quotes your own price

Two questions we were cited on one week, we lost the next. The log showed why. The answer was now built from independent pricing guides that listed our competitors’ prices and not ours. Our pricing page was public the whole time. The directories simply had not added us, and the assistant trusted the aggregator over the vendor.

That moved a task from the bottom of our list to the top: get listed, with current figures, in the guides the assistant already reads. It is outreach work rather than content work, and the same weekly log tells you whether it worked.

4. The sources column tells you where your market lives

For some of our questions the answers were built from vendor pages every single week. For others, from a rotating set of comparison sites and pricing directories. 

If your category’s answers come from vendor pages, publish better vendor pages. If they come from directories, be in the directories. The log tells you which, and it costs an hour a week.

What did not move the needle

  • Adding “verified on” dates to our fact blocks. The right thing to do, and no measurable effect on its own.
  • Structured data. We ship it. The assistant cites pages that lack it just as often.
  • Traffic as a proxy. Our search clicks were flat while the assistant answers moved. AI visibility and search traffic are different curves. Measure them separately or you will conclude nothing.

How to run this for a client in an afternoon

  1. Write ten to fifteen questions in the buyer’s words. Include a few that describe the client’s proposition without naming the brand.
  2. Ask them every week in a clean session with web search on. Log cited, recommended, and the source websites.
  3. Fix the words first (titles, opening paragraphs). Then the formats (lists before head to head pages). Then the third parties (the sites in the source column).
  4. Report it next to the search numbers, never blended with them.

The log itself is a spreadsheet. The hard part is doing it every Monday. If you want to see what the pages in this story look like, one of the roundups and our pricing page are the ones that did the work.