Question 1

What does Canary actually do?

Accepted Answer

Canary runs AI behavioral friction audits on conversion-focused pages. Five deeply-architected customer personas walk your product across three frontier AI systems running independent passes. Findings only graduate when the runs converge. Single-model signals are quarantined for human review. Rob Bulford, the Behavioral Architect, reviews and signs every report we sell — Standard, Funnel Bundle, and Deep-Dive.

Question 2

How is this different from a generic AI 'audit' tool?

Accepted Answer

Three real differences: (1) we run three independent AI systems and only ship findings that survive consensus and adversarial review (most tools are single-model heuristic scans); (2) every paid report is read line-by-line and signed by a human Behavioral Architect (most tools have no human in the loop); (3) every audit is pre-registered to a public ledger before it runs, blocking p-hacking. This is a real behavioral study with real methodology, not a fly-by-night AI wrapper.

Question 3

How is this different from a traditional UX research firm?

Accepted Answer

Speed and price. Traditional UX research projects average 42 days from kickoff to deliverable (Dscout, 2024, survey of 300+ UX researchers) and cost a median $18,000 (Rosenfeld Media, 2024). Canary delivers a Standard audit in 48 hours for $249. We're not a replacement for in-depth human research. We're the diagnostic layer that catches friction in the funnel quickly and cheaply.

Question 4

Who is Canary built for?

Accepted Answer

Direct-to-consumer (DTC) e-commerce brands at $1M–$50M revenue. The engine is purpose-built for pages that drive sales: homepage, product page, cart, checkout, signup. Operators who own the funnel and can implement fixes quickly get the most value. The methodology can be applied to other digital products with conversion-driven flows, but DTC is where it's been most empirically validated.

Question 5

What is the Triple-AI Consensus Engine?

Accepted Answer

Three frontier AI systems (Anthropic's Claude, OpenAI's GPT, and Google's Gemini) independently walk your product through five customer personas. Findings are then merged via a separate Consensus Comparison step (run on a different model family to avoid self-grading bias). Findings all three agreed on lead the report. Findings two of three surfaced are included with the dissenting model's reasoning noted. Single-model signals are quarantined for human review.

Question 6

What are the nine pillars of rigor?

Accepted Answer

Heterogeneous Population (5 distinct buyer profiles, not one), Pre-Registration (hypothesis published before the audit runs), Multi-Model Consensus (three frontier AI systems agree), Literature Anchoring (findings tied to published behavioral research), Transparency (methods, scope, and limits are public), Adversarial Validation (findings are red-teamed before delivery), Human Sign-Off (every paid report is architect-signed), Calibration (every audit measures itself against known patterns), Known Limits (we publish what AI personas cannot measure).

Question 7

What is pre-registration?

Accepted Answer

Before any AI persona walks your page, we publish the hypothesis, persona distribution, and success criteria to a public ledger. This blocks p-hacking. We can't reframe what we found to fit a story after the fact. The pre-registration record for every paid audit is verifiable on canary-research.com.

Question 8

What is Adversarial Critique?

Accepted Answer

After the three AI systems produce their findings, the engine runs a separate red-teaming pass: Claude reviews every finding and tags it Keep / Revise / Drop based on eight failure-mode categories (hallucination, generic-vs-specific, severity inflation, causal overreach, etc.). Findings that don't survive critique are revised or dropped before the report builds.

Question 9

How are the personas designed?

Accepted Answer

Five customer personas are deeply architected with distinct cognitive bias profiles, device contexts, prior shopping patterns, and purchase goals. Persona 01 is a comparison-shopper, P02 is trust-anxious, P03 is subscription-resistant, P04 is mobile-driven, P05 is bulk-order skeptical. Persona variance comes from prompt architecture, not from rotating models. The same five personas run across all three AI systems for true cross-model consensus.

Question 10

How long does an audit take?

Accepted Answer

The Standard Audit ($249, human-verified) is guaranteed within 48 hours. The Funnel Bundle ($549, three Standards) delivers within 72 hours. Deep-Dive Studies (custom scope) run 7–10 business days because of the additional flows, custom persona engineering, and deeper architect review.

Question 11

What does the architect review add over the raw engine output?

Accepted Answer

Same engine, different filter and review depth. The engine alone ships only Consensus 2/3+ findings without human review — that's what the free Friction Score scan gives you a taste of. The Standard Audit includes every finding the engine catches, with a Behavioral Architect's line-by-line review: weak findings cut, severity recalibrated, and architect-verified single-signal findings included. There is no quota — the count reflects what the page actually contains. The review also catches real-world errors that the engine can occasionally produce (like temporal-context errors or hallucinated specifics) that automated review can't catch.

Question 12

What does each audit cover?

Accepted Answer

Each Standard audit covers ONE public-facing page per audit run. Pick any page (homepage, pricing, signup, checkout, product, account, or any other public page). The Funnel Bundle is three Standards bundled. Pick any three pages, save $198 vs buying individually. Deep-Dive Study covers multiple pages in one engagement, including behind-the-login flows with test credentials you provide.

Question 13

Can Canary audit pages behind a login?

Accepted Answer

Yes, but only on a Deep-Dive Study (custom scope, quoted after a short brief). With test credentials you provide, we audit authenticated screens, dashboards, multi-step wizards, and CSV upload paths. API integration tests and connector flows are out of scope for AI personas (those need real users). We tell you up front what AI personas can and can't test on your specific product before you buy.

Question 14

What's the refund policy?

Accepted Answer

Full refund within 7 days of delivery if the findings are not actionable. We define 'actionable' up front in the kickoff brief. If the report doesn't meet that bar, you get your money back. The Funnel Bundle has the same actionability guarantee, applied per audit.

Question 15

How do I pay?

Accepted Answer

Standard Audit and Funnel Bundle are paid by card at checkout through Stripe — click the tier, pay, and the audit kicks off. Deep-Dive Studies are quoted first: scope is confirmed in writing, the pre-registration record is published to the public ledger, then we invoice (credit card or Interac e-transfer). Prices listed in USD.

Question 16

What can AI personas NOT measure?

Accepted Answer

Five things we publish openly: (1) emotional and visceral brand response (AI personas don't have a body); (2) FDA / HIPAA / clinical-grade claims (out of scope, requires real human subjects); (3) accessibility audits with assistive-tech users (AI personas catch obvious patterns but can't replace screen-reader testing with real users); (4) email lifecycle and post-purchase flows (out of scope for page-based audits); (5) deep performance / Core Web Vitals beyond what's visually observable (for that, use Google Lighthouse or a paid Real User Monitoring tool).

Question 17

Why don't you do dark-pattern optimization?

Accepted Answer

We refuse projects that ask us to optimize for psychological manipulation: pre-checked subscriptions designed to escape notice, fake urgency, manipulative friction. Behavioral science is a tool that works in both directions; we use it to reduce friction for buyers, not to extract revenue against their interests. This is non-negotiable and published in our Brand Guide.

Question 18

Have you been wrong before?

Accepted Answer

Yes. We audited a client and the engine produced a high-severity finding flagging 'Future-Dated Content' because the AI models' training data assumed the current year was earlier than it actually was, but the site correctly displayed the current year. The Behavioral Architect caught it during review and dropped the finding before delivery. That's exactly why every report we sell carries an architect review. We've since fixed the prompts to inject today's actual date into every audit.

Question 19

How do I verify your claims?

Accepted Answer

Two independent paths today, three once the Pre-Registration Ledger fills with paid audits: (1) the published whitepaper documents the methodology in full; (2) the self-audit on this site is real, you can read it, verify the findings against the live page, and see exactly what a paying client receives.

Question 20

What happens if your engine misses something important?

Accepted Answer

Honest answer: it can. AI personas have known limits (see above), and AI models have training-cutoff and context limitations. The architect review on every paid report is specifically designed to catch errors of this type before delivery. The free Friction Score scan ships automated output without human review — quick directional input rather than client-shippable rigor, which is why it's free.

Your website is quietly costing you sales every single day.

What's your site's Friction Score?

Score a page

Snapshot of yourstore.com/checkout

Shipping cost only revealed at credit card entry

Your funnel is leaking revenue.
You don't know where.

From kickoff to delivered report in 48 hours.

Why the findings hold up.

Who reviews your report

The methodology is published.

Transparent, flat-rate pricing.

Standard Audit

Funnel Bundle

Deep-Dive Study

We publish what we won't do.

What Canary Does

What Canary Refuses

FAQ

Chapter 1 · What is Canary?

Chapter 2 · How does it work?

Chapter 3 · Pricing & deliverables

Chapter 4 · What we won't do

Request an Audit.