Building an AI-search baseline across a global automotive portfolio

The challenge was consistency, not another visibility score

A global automotive group needed to understand how its brands and products appeared across AI-powered discovery. The scale made one-off prompt checks almost useless: each market, brand and customer journey could produce a different story.

The useful question was not simply whether a brand appeared. It was how accurately it was represented, when it entered a recommendation set, which competitors appeared beside it and what sources seemed to support the answer.

I designed a framework that could survive contact with a large organisation

I helped turn an emerging discipline into a documented research programme: agreed journeys, a controlled prompt sample, consistent capture conditions and separate measures for presence, recommendation, citation and message accuracy.

The work drew together specialist AI-search tooling, manual investigation and traditional technical and content evidence. That combination mattered because automated scores can reveal a pattern, but rarely explain what a team should do next.

• Comparable prompt and journey sets across brands
• Citation and source-pattern analysis
• Entity, product and competitor representation checks
• Prioritised technical, content and authority recommendations

The measurement problem had several moving parts

The work covered brands with different products, markets, audiences and levels of existing authority. Answers could also vary between models, sessions and prompt wording. A single percentage could hide all of those differences and create confidence that the underlying evidence did not deserve.

I separated the stable parts of the programme from the exploratory ones. Core journeys and prompts created continuity. Additional tests were used to investigate emerging questions without quietly changing the baseline every time the research evolved.

Tools accelerated collection; judgement made the findings useful

Specialist platforms including AccuLLM and AccuRanker helped collect and organise a large volume of observations. Manual review then tested the meaning behind the numbers: whether a mention was genuinely useful, whether a citation supported the claim being made and whether a competitor appeared for a reason the organisation could act on.

I looked for patterns across journey, market, brand, model and source type. That made it possible to distinguish isolated output variation from a repeated visibility or representation problem. Every material finding retained a route back to the prompt, answer and evidence that produced it.

The output had to work for specialists and decision-makers

Findings were translated into detailed working decks and concise stakeholder narratives. Each recommendation connected an observed problem to the team able to act on it, with the evidence and limitations left visible.

That made the baseline more than a report. It became an operating model the organisation could rerun after changes, compare across brands and refine as AI-search behaviour and measurement tools developed.

The result was a baseline the account could build on

The programme gave the account a common language for a fast-moving subject and a defensible starting point for investment. Instead of chasing screenshots, teams could see where visibility problems originated, which brands or journeys needed deeper investigation and which changes were worth testing first.

Because the method, limitations and evidence were documented, later measurement can use the same frame of reference. That is essential for understanding whether a change coincided with real movement or whether the apparent improvement is simply normal output variation.

What I would carry into the next enterprise GEO programme

AI-search measurement works best when it is treated as research rather than rank tracking with a new label. The strongest programmes keep a controlled baseline, preserve the underlying answers and allow room for qualitative investigation.

The other lesson is organisational. A useful finding must reach the team capable of changing the evidence environment, whether that is technical SEO, content, digital PR, product, brand or local-market ownership. Measurement only becomes valuable when it changes a decision.