Aiso field report · 3 engagements
What actually changes AI recommendations?
First published .
| Brand | Work | What we can say |
|---|---|---|
| Particle Public outcome | Real-prompt research, crawlability, schema, content/structure work, off-site source strategy, attribution | Approximately 5x AI visibility during the engagement, publicly reported by mind Retail |
| Sophia High School Ongoing + renewed | Prompt mapping, technical review, llms.txt, schema, third-party analysis, 100+ content ideas, redesign, bot dashboard | Renewed after the initial engagement; live measurement continues |
| Stay Unique Ongoing | B2B + B2C audits, booking-app review, schema, listing clarity, third-party benchmarking, attribution | Technical and listing sprints completed; reporting continues by city and prompt |
1. Retrieval problems have to be fixed before persuasion problems
Across all three engagements, some of the first work was boring: crawlability, page structure, schema, canonical URLs, listing URLs, or making the right facts available on the right surface. That is not because schema is magic. It is because a model cannot ground an answer in evidence it cannot reliably reach or interpret.
My rule now is simple: if a brand loses because the evidence is missing or unreachable, do not start with more “GEO content.” Fix the evidence path.
2. The brand website is only one part of the answer supply
Particle and Sophia both required explicit third-party source analysis. Stay Unique adds another version of the same problem because OTAs, direct listings and other travel surfaces describe the same property. When assistants lean on external sources, an immaculate homepage does not control the whole answer.
This is the part traditional on-site SEO plans routinely underweight. The question is not just “what do we say?” It is “which pages does the model trust when it checks?”
3. Real prompts are better briefs than keyword variants
For all three brands, the useful unit was the buyer situation: dandruff or hair thinning for Particle, parent constraints for Sophia, city/amenity/traveler combinations for Stay Unique. Those prompts reveal the conditions under which a recommendation is made.
That changes the content brief. Instead of producing another broad category article, we can see the missing decision criterion, the competitor already satisfying it, and often the fan-out search the assistant uses to resolve it.
4. Multi-surface consistency matters more as the purchase gets local or inventory-heavy
Stay Unique made this unusually visible. The same property can exist on a consumer site, a B2B site, a booking application and third-party channels. If names, attributes, descriptions or URLs drift, the assistant has to reconcile conflicting evidence before it can recommend anything.
That is why we built Listing Clarity work into the engagement rather than treating listings as a separate local-SEO chore.
5. A visibility score alone cannot tell you what worked
The public Particle outcome is useful, but it is also a warning. A fivefold visibility lift across a combined intervention does not identify the causal contribution of schema versus content versus third-party work. Sophia and Stay Unique therefore track a wider chain: prompts, mentions, citations, bot activity, AI-referred sessions and conversions where possible.
The practical lesson from the three engagements is not a universal ranking factor. It is a debugging order: make the evidence reachable, make it credible in the sources the model checks, answer the actual buyer constraint, then measure whether the recommendation changed.