The first version of AskRally let you ask 100 AI-generated personas for feedback on your ideas. Today, we’re changing the default to 5 personas.
Why? The quality of responses has improved dramatically since we started training our personas on real humans (our GenPop™ panel). Now, you only need to read a handful of responses to understand what’s great—or terrible—about your idea.
A New, Qualitative-First Experience
We’ve redesigned AskRally to show individual responses first, instead of aggregate summaries.
Nobody wants to read 100 comments at once, so sampling (5–20 personas from a larger audience) is now the default.
This redesign also allows us to support upcoming features like video reactions from individual personas.
If you already have a paid subscription, nothing changes.
You can stay on your current plan until you choose to cancel or upgrade.
For new users, we’ve launched updated plan options.
Why This Matters
This shift aligns with how power users are using AskRally today—co-creating ideas with realistic personas. Virtual focus groups help you:
-
eliminate bad ideas quickly
-
iterate on good ones
-
shoot your best shot in the market
This qualitative focus doesn’t mean we’re abandoning quant. We simply believe quant only matters when the personas are grounded in real people. Our panel is only 200 members right now (100 added this month!), and expanding it will make quant useful again soon.
For now, quant services are available for enterprise clients who want to build custom panels.
👉 Talk to sales.
Storytime:
How we got here and decided to refocus the company away from quant polling towards qual with persona-level models is an interesting one. We think there are a lot of implications for the synthetic research industry that will be instructive to understand.
One of the early wins we had at AskRally was optimizing an AI audience until it could predict the result of the last election with 98% accuracy:
It was interesting academically that we removed political correctness bias (personas voted 90% for Harris without calibration), but there was an underlying problem that terrified us.
When we looked at the individual level the responses sounded like AI, not how real people speak, and there was no consistency with their profiles. We had hardcore republicans voting blue and the most liberal personas voting red. The numbers added up correctly in aggregate, but hiding underneath those headline numbers the individual persona models were wrong. Turns out near-perfect accuracy on a coin flip isn't impressive, it just means your errors are canceling each other out.
Most vendors stopped there and declared victory. Next time your pitched on accuracy numbers, ask them to lift the hood. You'll quickly find it's not even easy to define what 'accuracy' means at the individual level, and when you do figure it out, you often realize the metric was designed to hide the coin flip.
So we spent the past six months solving the right problem. Which led us to building our custom panel, GenPop, based on interviews with hundreds of real people. We saw a huge quality boost by building a persona model for each individual person, training it until it’s smart enough to trick a superhuman judge into thinking it’s human.
Can you tell which one is the human response below?
The quality has jumped so high that we decided to stop hiding those responses behind a summary, and instead display all the individual persona responses right there on the page.
You don't need to browse 100 different responses anymore when you can ask 5 personas and get quick directional feedback before iterating. Because the personas are grounded in real people, they give specific, surprising pushback that aligns with real-world feedback, not generic AI responses that all sound the same.
The better question with AI market research is "Is this useful?".
Most synthetic studies rely on the "wisdom of the crowds" effect, query enough personas and you'll replicate aggregate results to 85% accuracy. But when you drill in, the individual responses don't align with what that profile would really say, and they sound like AI, not real people.
To fix this, we built our own panel from scratch: real humans, 30-minute cheat-proof video interviews, and our proprietary calibration engine training each persona until it's indistinguishable from the real person. The result? You get the same diverse, counterintuitive reactions you'd hear in real focus groups, but you can iterate in real-time for next to nothing.
This was originally an enterprise-only service, but the quality jump was too good to gate. So we've reworked our plans to make GenPop the foundation for everyone. If you're already subscribed, nothing changes unless you want to upgrade.
Smarter Sampling
AskRally now offers multiple ways to build virtual focus groups:
-
Random sampling – Choose personas randomly from GenPop
-
Manual sampling – Hand-pick personas and filter by keyword
-
Relevance sampling (coming soon) – Vector search selects relevant personas automatically
Ask about retirement? Expect personas age 55+.
The New Workflow
Everything is designed to be iterative and insight-driven:
-
Submit your ideas or questions
-
Review individual responses (generate more on demand)
-
Filter out bad ideas, refine promising ones
-
Generate video reactions when needed
Need Quant Research?
If you need large sample sizes for statistical validation, we now offer a dedicated quant option. We can help you:
-
clone personas anchored to real humans
-
expand GenPop or build a private panel
-
verify demographic distributions
-
build automated integrations (e.g., fan-out testing)
👉 For enterprise-scale quant, contact sales or reply to this email.
More Good News
-
GenPop just doubled — 100 new personas added (200 total and growing monthly)
-
Calibration API — coming soon for your own workflows
-
Response quality upgrade — Google model now the default (previously smart-plan only)
We believe these changes will help you get better insights, faster.
Questions or concerns?
Just reply to this email.
Thank you for being part of our journey.
Best regards,
Team AskRally