Prompts now run from the perspective of the buyer you're targeting, not a generic anonymous user.
What's new
- Attach one persona to any monitor. Every prompt in that monitor runs as if asked by that specific person.
- Build the persona with AI or write it yourself. Either way it describes a role, a situation, constraints, priorities, and what they weigh before deciding, not a script for how the model should write.
- Enable stacking when you create the monitor, or turn it on later in settings.
- Every response now shows which persona asked the question, right in the response details view, next to the model, the sentiment, and the score.
- Personas in use are protected. Detaching one asks for confirmation first, and a persona tied to tracked responses can't be deleted.
Why it matters
A blended score tells you if you're visible. It doesn't tell you if you're visible to the buyer who's about to make a decision, and whether the model is recommending you or quietly steering around you.
For B2B SaaS for example, the same product gets researched by completely different roles at completely different moments in a deal, and each one asks a different question:
- A technical evaluator doing due diligence asks about API rate limits, SOC 2 status, self-hosted vs. cloud, and migration effort, not "best tool for X."
- A department head comparing options asks about pricing tiers, seat limits, and time-to-value, closer to a commercial-intent prompt than a technical one.
- An economic buyer signing the contract asks about total cost of ownership, vendor stability, and what happens if the vendor gets acquired or shuts down a feature, questions a generalist prompt never surfaces.
- A first-time evaluator with no category experience asks a much more basic "what does this category of tool even do," and the model tends to answer with the most recognizable name, not necessarily the best fit.
A blended visibility score averages all four of those into one number. It can tell you you're visible without telling you that you're the top pick for procurement teams doing security due diligence, but an afterthought when a CTO is comparing architecture, or the reverse. Intent type already separates branded, informational, commercial, and transactional prompts. Persona stacking adds the missing variable: not just what stage of the funnel someone's in, but who they are and what they're optimizing for, since a security-focused buyer and a speed-focused buyer read the exact same category very differently.
That's the gap between "we rank well in AI search" and "we show up, in the right role, for the specific person about to decide."
How it works
Each persona gets a dedicated stacking prompt, the description you wrote or generated. When stacking is on, that prompt is injected around the tracked prompt at run time, so the model answers the way it would for a known user with history, not a first-time anonymous query.
The prompt deliberately stays out of instructing the model on tone or format. It states who's asking and what they care about. No "act as," no brand mentioned, no house style baked in. That keeps the result a read on how the model treats that buyer, not a read on how well you wrote the instruction.
Response details now carry the persona used for that run, so when a score moves, or when your position in the answer shifts, you can check whether it moved for everyone or just for one segment.
Get started
Persona stacking is live now on every plan. One persona per monitor. Add a persona of at least 30 characters in monitor settings, flip on stacking, and the next scheduled run picks it up.
