Every AI Visibility Success Story Starts With the Same Step. Most Businesses Skip It.

Four businesses. A car parts retailer, a three-person consultancy, a travel insurance marketplace, a headwear brand. Different industries, different sizes, different tactics.

Read their stories side by side and the same thing sits at the start of every one.

Before anyone wrote a page or pitched a mention, they measured. The parts retailer wrote out a hundred commercial prompts and tested them daily on three platforms, then checked its post-purchase survey to see how many customers named an AI assistant. The consultancy knew it stood at 7% before it knew why. The travel marketplace surveyed nearly 2,500 customers and learned that 81% found AI helpful when shopping for cover. The headwear brand's own specialist said the team had been blind until it built a way to see.

None of them started with content. They started with a number.

Why most businesses do not

Measuring AI visibility feels optional in a way that measuring website traffic does not. There is no analytics dashboard that comes free with the domain. And the first result is usually uncomfortable.

Consider what the broader data says. A 2026 index that analysed 70 B2B companies found only 4.3% maintained a healthy discovery funnel where their brand appeared in early-stage buyer questions. Put another way, roughly 96% were invisible at the moment a buyer first asked an assistant for options. That is not a niche problem. It is the default.

Meanwhile, the traffic is arriving. Visits to B2B brands referred by ChatGPT rose from about 645,000 in June 2025 to 2.6 million in June 2026, a 303% increase, according to Demandbase data cited by Digiday. And 58% of marketers say AI-referred visitors convert at higher rates than traditional organic traffic, per HubSpot's 2026 State of Marketing report.

So the channel is growing, the visitors are valuable, and most businesses have never checked whether they exist in it.

What a real baseline looks like

Here is where a lot of businesses measure wrong, and it is worth being exact, because a bad baseline is worse than none.

The largest public study of AI recommendation consistency had 600 volunteers run 12 prompts a combined 2,961 times across ChatGPT, Claude and Google's AI. Fewer than one in a hundred pairs of answers contained the same list of brands. One check tells you almost nothing. The same study found that visibility measured as a rate across many runs is a reasonable metric, and that a "rank position" in AI is not a real quantity at all.

So a baseline is not "I asked ChatGPT once and we came up." It is:

The prompts your buyers actually type. Not your keywords. The questions, in natural language, that end in a purchase in your category. Ten to twenty of them is a start.

Run repeatedly. The same prompt, many times, so you get a rate rather than a lucky draw.

On the assistants your customers use, and on the free tiers as well as paid ones, because those are different products with different models and fewer links shown.

From the places your customers live. Location changes the answer. A prompt run from a data centre is not the prompt your customer ran from their kitchen.

Recording who gets named, not just whether you do. Your competitors' share is half the picture.

Then, if you can, ask your customers how they found you. The car parts retailer's survey result, from under 0.5% to 5% naming AI, is the most credible figure in any of these stories, because the customers said it themselves.

What the baseline buys you

Three things.

A reason. It is hard to fund work against an invisible problem. A sentence like "we appear in a fraction of the prompts our buyers use and our nearest competitor appears in most of them" is one a budget can be built on.

A direction. The prompts where you are absent and a competitor is present are the work list. The parts retailer's brand-mention push, the consultancy's structural fixes and the headwear brand's prompt portfolio all came from looking at a baseline and seeing where the gaps were.

Proof. Six months later, the only way to know the work did anything is to compare against where you started.

The uncomfortable part

For most businesses the first baseline shows a small number. That is the expected result, not a failure. Every company in this series started small: 1%, 7%, single digits.

The businesses that grew were not the ones with the best first number. They were the ones that took it.

References

1. 2X AI Innovation Lab, "2026 AI Visibility Index," via GlobeNewswire, 7 April 2026. globenewswire.com
2. InformaBTL, citing Digiday and Demandbase data on ChatGPT-referred B2B visits. informabtl.com
3. Fishkin, R. and O'Donnell, P., SparkToro, "AIs are highly inconsistent when recommending brands or products," January 2026. sparktoro.com
4. Press release via PR Newswire on the auto parts retailer's results, 18 May 2026. prnewswire.com
5. Squaremouth press room, 21 October 2025. squaremouth.com
6. Am I Cited, "AI Search Visibility Revenue: 6 Case Studies," including the HubSpot 2026 State of Marketing statistic. amicited.com