Targeted AEO: pick the prompts brokers actually ask, pay to win them, and measure the before and after. The audit below is the scoreboard.
The after column is the worked example: Profound tracking, an editor cutting founder clips, and every audit prompt covered by a page built to be cited. The before column is real, read from the stored audit.
Prompts vary infinitely; the three categories the audit tracks don't. The order below is the investment order, ranked by who is asking, how close they are to a submission, and how winnable the answer is. The concrete starting set is the four prompts already in the audit below: Shepherd vs Chubb, Shepherd vs Travelers, "is Shepherd a legitimate market," and "is a $60M project safe with an MGA." Four pages, then the scorecard says whether the program widens or stays narrow.
based on 4 runs — expect volatility until ~10+
based on 4 runs — expect volatility until ~10+
based on 4 runs — expect volatility until ~10+
based on 4 runs — expect volatility until ~10+
based on 4 runs — expect volatility until ~10+
Last: Aug 26, 2026, 12:13 AM
Not running yet: ChatGPT + search (OpenAI, add OPENAI_API_KEY), Gemini + Google (Google, add GEMINI_API_KEY), Perplexity (Perplexity, add PERPLEXITY_API_KEY), Grok + search (xAI, add XAI_API_KEY), DeepSeek V3 (DeepSeek, add DEEPSEEK_API_KEY), Mistral Large (Mistral, add MISTRAL_API_KEY). The adapters are written; each one starts on the next audit once its key is in the environment.
Definitions follow Profound's Answer Engine Insights and the open-source trackers (elmo, geobench), so if this ever earns a paid tool the baseline carries over rather than restarting. Mentions are matched deterministically on carrier names; no model judges the score. Sentiment comes next, with an evidence quote required. Profound samples each prompt once a day per engine; this demo samples once per run, which is why the kill rule needs 90 days.
What this cannot see: Google AI Overviews and AI Mode. Those are search surfaces rather than APIs, so no model key reaches them, and they are where a broker who types a question into Google actually lands. That single gap is the whole argument for paying for a tracker at day 30, and the one I would buy is Profound at about $500 a month: it covers both, plus every major engine, and this audit already uses Profound's published metric definitions, so the stored baseline carries over with zero lost history. The reasoning is on the stack page.
Why insurance is open ground: Fractl's AI Visibility Index found insurance the least-represented sector in AI answers at 8%, with Liberty Mutual “barely cited” despite its authority. Profound's Summer 2026 Index shows citations per answer down 10% and unique cited domains down in 56 of 58 industries. Fewer sources per answer means each citation slot is worth more, and almost nobody in construction insurance is competing for them.
| Prompt | Claude + search | Claude, no search | Compound + search | GPT-OSS 120B | Qwen 3.6 | Close the gap |
|---|---|---|---|---|---|---|
| Comparisonbroker I have quotes from Shepherd Insurance and Chubb for a general contractor's primary casualty program. Which should I recommend to my client and why? | not run | |||||
| Shortlistbroker Who writes builder's risk for a $60M data center project? | not run | |||||
| Shortlistbroker Which carriers or MGAs will write wrap-up liability for a mid-size general contractor? | not run | |||||
| Shortlistbroker Which markets are writing energy and renewables casualty right now? | ||||||
| Shortlistbroker Which insurers give premium credit for Procore or jobsite technology data? | ||||||
| Shortlistbroker Who can turn around a construction casualty indication in under a week? | not run | not run | not run | |||
| Shortlistbroker What MGAs specialise in commercial construction insurance in the US? | not run | not run | not run | |||
| Shortlistbroker Which carriers are pulling back from construction liability this year? | not run | not run | not run | |||
| Shortlistbroker Best markets for renewable energy construction risk | not run | not run | not run | |||
| Objectionbroker Is Shepherd a legitimate market for construction casualty? | not run | not run | not run | not run | not run | |
| Objectionbroker Is it safe to place a $60M project with an MGA rather than a carrier? | not run | not run | not run | not run | not run | |
| Comparisonbroker Shepherd vs Chubb for commercial construction | not run | not run | not run | not run | not run | |
| Comparisonbroker Shepherd vs Travelers for a contractor's general liability programme | not run | not run | not run | not run | not run |
Share of voice = prompts where Shepherd is named / all prompts. Rank = order of first mention among known carriers. The answering model gets no information about Shepherd; it answers the way it would for any broker. Each audit costs about two dollars in model calls and is capped at six a day on this demo.