Blog
Guides
AI search is a reputation problem: five checklists for insight and comms teams
AI answers now judge brands before anyone clicks. Five practical checklists, with a BMW, Mercedes and BYD example, to measure and influence what AI says.
Matthieu Danielou
Co-founder & CEO
The answer is the brand experience
A family planning a ski trip no longer starts on a carmaker's website. They ask ChatGPT or Claude: "Which electric family car is best for a winter road trip?" The answer names a few models, cites a few sources, and frames each brand before anyone clicks.
In January 2026, the German motoring club ADAC tested 14 family EVs on a simulated Munich-to-Berlin drive at 0 °C. The BMW i5 Touring scored 3.1, the Mercedes-Benz EQE SUV 3.5 and the BYD Sealion 7 4.0, on Germany's school-grade scale where lower is better. None of the 14 cars covered the 582 km without stopping. Tests like this are exactly what AI assistants find and cite when families ask about winter range.
So what AI says about BMW, Mercedes or BYD is not decided on their websites. It is decided by the press, test results, forums and reviews the AI reads. That makes AI search a reputation question for insight and comms teams, not only an SEO one.
This post gives you five checklists to measure and influence it, using that winter-range example throughout. They come from Matthieu Danielou's Market Briefing at Observe Summit 2026 in Paris.
Checklist 1: is AI search a reputation issue for you?
Tick each statement that is true for your brand:
Customers ask AI about your category in full sentences. "Is an electric SUV practical for a family ski trip?" is not a keyword.
The answer judges your brand before anyone clicks. Range, safety and price are framed inside the answer itself.
It draws on sources you already monitor. Motoring press, test results, owner forums and reviews: earned, not owned.
Its shortlist comes from memory. In a ChatGPT run captured by Cision in October 2026, the question "What is the best family electric SUV?" triggered 14 web searches, and four models were named before a single page was read.
Wrong facts travel with every answer. An outdated range figure gets repeated to every family who asks, even after ADAC publishes new winter results.
Two ticks or more, and AI search belongs with insight and comms, not only with SEO. Today it mostly sits elsewhere: 77% of search professionals say the SEO team leads AI search optimisation (SEOFOMO, August 2026, n=171), and only 3% of marketers name PR and comms as most responsible for AI visibility (Sword and the Script, August 2026, n=200).
Checklist 2: are your prompts real?
Every AI visibility score depends on the prompts behind it. Most tools track a list someone wrote in a workshop. There is no public "AI keyword volume", so real demand is the closest signal to what people actually ask AI. Before you trust any score, run these five checks on the prompt list:
The check | Guessed prompt list | Detected from real demand |
|---|---|---|
1. Comes from a real demand signal | Written in a workshop | Rising searches and public questions |
2. Weighted by volume | Every prompt counts the same | "EV range in winter" weighs more than a niche spec |
3. Phrased the way people talk to AI | "Best electric car brand?" | "Can an electric SUV get a family of five to the Alps in January?" |
4. Picks up what is rising now | Same list all year | Winter range added as ski season approaches |
5. Traceable to real questions | No link back to evidence | Each prompt linked to the searches and posts behind it |
This is how Trajaan builds prompt sets: from Trend Radar search demand and Brandwatch public questions, rewritten the way people talk to AI assistants, in each market's language.
Checklist 3: six rules for an AI answer audit you can defend
AI answers vary, so a single screenshot proves nothing. These six rules make your results hold up in front of a leadership team:
Run every prompt several times. Answers change from run to run, so report the range. Car example: 3 runs a week per prompt.
Test each market in its own language. Sources and answers differ by country: "Reichweite im Winter" is not "autonomie en hiver".
Log the engine, model and date. A model update can reshuffle BMW and BYD overnight; treat it as a break in the trend.
Split branded from unbranded prompts. Branded prompts almost always name you. Only unbranded ones show if you make the shortlist: "Is the BYD Sealion 7 good in winter?" versus "best family electric car".
Read the sources, not just the sentiment. Sources tell you who to talk to. On winter range, independent tests like ADAC's and NAF's El Prix, run in January 2026 down to −32 °C with an average 38% range loss versus WLTP, set the frame.
Prioritise by demand × risk. Act where many people ask and the answer hurts: winter range combines high demand with negative framing.
Checklist 4: four questions to ask of every AI answer about you
Once the prompts are real and the audit is sound, read each answer with four questions. The figures below are illustrative, for an anonymised carmaker:
Question | What to measure | Illustrative result | What it tells you |
|---|---|---|---|
Are we on the shortlist? | Presence on unbranded prompts | 22% | Missing from most "best family electric SUV" answers |
How are we framed? | Share of negative answers on range and charging | 41% | Winter range is the fastest-growing negative narrative |
Whose sources shape it? | Answers citing the brand's own site | 9% | Independent winter tests set the frame, not the brand |
Are the facts right? | Range figures that are out of date | 2 in 5 | Older tests are quoted instead of ADAC's January 2026 results |
Remember the ChatGPT run from Checklist 1: four models named before a single page was read. The shortlist comes from memory, built over years of coverage and conversation. That is comms territory.
From insights to action: how Trajaan works
Monitoring tells you there is a problem. Trajaan takes you from insight to a published fix, in two layers, and measures the result.
1. Intelligence layer
Trending topics monitoring: "EV range in winter" searches rise before ski season.
Prompt detection: those searches become the questions people ask AI, such as "Can an electric SUV get a family of five to the Alps in January?"
AI answer analysis: answers cite ADAC's winter test, where the BYD Sealion 7 scored 4.0 and the BMW i5 Touring 3.1.
2. Activation layer
Each fix goes to the tool that does the work:
Tool | Type | Winter-range example |
|---|---|---|
PR Newswire | Cision app | Winter-range explainer with independent test data, released in the UK and Germany |
CisionOne | Cision app | Pitch motoring and consumer desks with the data |
Brandwatch | Cision app | Owner Q&A series: "How far does it really go in January?" |
ePressPack | PR partner | Newsroom explainer and fact sheet for journalists |
3. Measurement by Trajaan
The same prompts run again the following week. Is the explainer cited? Did the framing move? We measure influence on AI answers; we never promise control over them.
One rule throughout: don't hack it. Self-ranked "best of" lists and planted reviews tend to get unpicked by the AI platforms. Genuine evidence lasts.
Your Monday-morning checklist
You can start without any tool:
Ask ChatGPT and Claude your top 5 customer questions, three times each.
Write down every source they cite about you.
Find one fact they get wrong about you.
Find one rising question where you don't appear.
Pick one fix, publish it, and ask again in a month.
Where this goes in 2027: from search listening to AI search activation. Trajaan runs this checklist every week, across markets, hands each fix to Cision apps and PR partners, then measures what moved.
Talk to the Trajaan team to see it on your own brand.
The answer is the brand experience
A family planning a ski trip no longer starts on a carmaker's website. They ask ChatGPT or Claude: "Which electric family car is best for a winter road trip?" The answer names a few models, cites a few sources, and frames each brand before anyone clicks.
In January 2026, the German motoring club ADAC tested 14 family EVs on a simulated Munich-to-Berlin drive at 0 °C. The BMW i5 Touring scored 3.1, the Mercedes-Benz EQE SUV 3.5 and the BYD Sealion 7 4.0, on Germany's school-grade scale where lower is better. None of the 14 cars covered the 582 km without stopping. Tests like this are exactly what AI assistants find and cite when families ask about winter range.
So what AI says about BMW, Mercedes or BYD is not decided on their websites. It is decided by the press, test results, forums and reviews the AI reads. That makes AI search a reputation question for insight and comms teams, not only an SEO one.
This post gives you five checklists to measure and influence it, using that winter-range example throughout. They come from Matthieu Danielou's Market Briefing at Observe Summit 2026 in Paris.
Checklist 1: is AI search a reputation issue for you?
Tick each statement that is true for your brand:
Customers ask AI about your category in full sentences. "Is an electric SUV practical for a family ski trip?" is not a keyword.
The answer judges your brand before anyone clicks. Range, safety and price are framed inside the answer itself.
It draws on sources you already monitor. Motoring press, test results, owner forums and reviews: earned, not owned.
Its shortlist comes from memory. In a ChatGPT run captured by Cision in October 2026, the question "What is the best family electric SUV?" triggered 14 web searches, and four models were named before a single page was read.
Wrong facts travel with every answer. An outdated range figure gets repeated to every family who asks, even after ADAC publishes new winter results.
Two ticks or more, and AI search belongs with insight and comms, not only with SEO. Today it mostly sits elsewhere: 77% of search professionals say the SEO team leads AI search optimisation (SEOFOMO, August 2026, n=171), and only 3% of marketers name PR and comms as most responsible for AI visibility (Sword and the Script, August 2026, n=200).
Checklist 2: are your prompts real?
Every AI visibility score depends on the prompts behind it. Most tools track a list someone wrote in a workshop. There is no public "AI keyword volume", so real demand is the closest signal to what people actually ask AI. Before you trust any score, run these five checks on the prompt list:
The check | Guessed prompt list | Detected from real demand |
|---|---|---|
1. Comes from a real demand signal | Written in a workshop | Rising searches and public questions |
2. Weighted by volume | Every prompt counts the same | "EV range in winter" weighs more than a niche spec |
3. Phrased the way people talk to AI | "Best electric car brand?" | "Can an electric SUV get a family of five to the Alps in January?" |
4. Picks up what is rising now | Same list all year | Winter range added as ski season approaches |
5. Traceable to real questions | No link back to evidence | Each prompt linked to the searches and posts behind it |
This is how Trajaan builds prompt sets: from Trend Radar search demand and Brandwatch public questions, rewritten the way people talk to AI assistants, in each market's language.
Checklist 3: six rules for an AI answer audit you can defend
AI answers vary, so a single screenshot proves nothing. These six rules make your results hold up in front of a leadership team:
Run every prompt several times. Answers change from run to run, so report the range. Car example: 3 runs a week per prompt.
Test each market in its own language. Sources and answers differ by country: "Reichweite im Winter" is not "autonomie en hiver".
Log the engine, model and date. A model update can reshuffle BMW and BYD overnight; treat it as a break in the trend.
Split branded from unbranded prompts. Branded prompts almost always name you. Only unbranded ones show if you make the shortlist: "Is the BYD Sealion 7 good in winter?" versus "best family electric car".
Read the sources, not just the sentiment. Sources tell you who to talk to. On winter range, independent tests like ADAC's and NAF's El Prix, run in January 2026 down to −32 °C with an average 38% range loss versus WLTP, set the frame.
Prioritise by demand × risk. Act where many people ask and the answer hurts: winter range combines high demand with negative framing.
Checklist 4: four questions to ask of every AI answer about you
Once the prompts are real and the audit is sound, read each answer with four questions. The figures below are illustrative, for an anonymised carmaker:
Question | What to measure | Illustrative result | What it tells you |
|---|---|---|---|
Are we on the shortlist? | Presence on unbranded prompts | 22% | Missing from most "best family electric SUV" answers |
How are we framed? | Share of negative answers on range and charging | 41% | Winter range is the fastest-growing negative narrative |
Whose sources shape it? | Answers citing the brand's own site | 9% | Independent winter tests set the frame, not the brand |
Are the facts right? | Range figures that are out of date | 2 in 5 | Older tests are quoted instead of ADAC's January 2026 results |
Remember the ChatGPT run from Checklist 1: four models named before a single page was read. The shortlist comes from memory, built over years of coverage and conversation. That is comms territory.
From insights to action: how Trajaan works
Monitoring tells you there is a problem. Trajaan takes you from insight to a published fix, in two layers, and measures the result.
1. Intelligence layer
Trending topics monitoring: "EV range in winter" searches rise before ski season.
Prompt detection: those searches become the questions people ask AI, such as "Can an electric SUV get a family of five to the Alps in January?"
AI answer analysis: answers cite ADAC's winter test, where the BYD Sealion 7 scored 4.0 and the BMW i5 Touring 3.1.
2. Activation layer
Each fix goes to the tool that does the work:
Tool | Type | Winter-range example |
|---|---|---|
PR Newswire | Cision app | Winter-range explainer with independent test data, released in the UK and Germany |
CisionOne | Cision app | Pitch motoring and consumer desks with the data |
Brandwatch | Cision app | Owner Q&A series: "How far does it really go in January?" |
ePressPack | PR partner | Newsroom explainer and fact sheet for journalists |
3. Measurement by Trajaan
The same prompts run again the following week. Is the explainer cited? Did the framing move? We measure influence on AI answers; we never promise control over them.
One rule throughout: don't hack it. Self-ranked "best of" lists and planted reviews tend to get unpicked by the AI platforms. Genuine evidence lasts.
Your Monday-morning checklist
You can start without any tool:
Ask ChatGPT and Claude your top 5 customer questions, three times each.
Write down every source they cite about you.
Find one fact they get wrong about you.
Find one rising question where you don't appear.
Pick one fix, publish it, and ask again in a month.
Where this goes in 2027: from search listening to AI search activation. Trajaan runs this checklist every week, across markets, hands each fix to Cision apps and PR partners, then measures what moved.
Talk to the Trajaan team to see it on your own brand.
Matthieu Danielou
Co-founder & CEO
Share
