Changelog

The old $4.99 unlock was a hangover from the public-beta era. It got people in the door, but it framed the product as the cheap version of HubSpot — which is free. We were anchoring against the wrong reference.

What changed

• Killed the $4.99 Stripe product. Existing $4.99 buyers grandfathered for life.
• New $49 AI Readiness Report — full consensus matrix, PDF, OG share card, competitor consensus, severity ranking.
• Free scan now shows enough to create urgency without satisfying it: AI Readiness Score, SEO-for-AI checklist, blurred consensus preview.
• $199 Full Accuracy Report adds per-claim accuracy scoring against verified site data and a correction playbook.

The full pricing matrix is documented in /pricing.html if you want to see how the tiers ladder.

Monitor ($399/mo) and Monitor Pro ($599/mo) both went live this morning. The infrastructure was the harder part: we moved scans into Cloudflare Workflows, added per-subscription nextScanAt bookkeeping, and built a separate drift-detection workflow that compares the latest scan to the prior one.

• Monthly re-scans (Monitor) and weekly re-scans (Pro), same brand each time.
• Email alerts when accuracy score moves by ≥ 5 points or a previously verified claim flips.
• Trend dashboard plots scores over time, per model and overall.
• AI Readiness Certification Badge — cached at CDN edge with 1-hour TTL.

The old report was a flat table. Everyone scrolled to the bottom for the score and ignored the rest. The new layout puts the score in a hero, surfaces the three most-disagreed-on claims first, and groups everything else by topic.

• Lead story: composite score + the one sentence that says why.
• Three-up: most-inaccurate, most-disagreed-on, most-recently-flipped.
• Grouped claim cards by dimension (positioning, pricing, founder bio, product list).
• Per-model reliability sidebar — at a glance, which model knows you best.

The free scan needed to deliver enough value to earn trust without satisfying the curiosity that drives the $49 purchase. We landed on five things: a composite readiness score, a pass/fail checklist, an AI Overview check, two copy-paste quick wins, and a blurred preview of the consensus matrix.

Everyone gets the file. The conversion play is the score below it.

One of the most common questions in support was "why does the AI think we're a B2C company?" Almost always, the answer was hidden in the rendered DOM — a stale meta tag, a JavaScript-rendered hero the scanner missed, or a robots block.

Every scan now captures the page exactly as the headless browser rendered it: the full DOM, a hero-viewport screenshot, and a full-page screenshot. They show up inline in the report and you can download them.

• Markdown — claim descriptions now render proper line breaks and emphasis in the consensus matrix.
• Hydration — the scan progress stepper no longer flashes a server/client mismatch on slow connections.
• Accessibility — corrected the aria-label on the homepage scan form.
• Copy — landing page now says "3 models queried" (ChatGPT, Gemini, Perplexity). Claude is on the roadmap, not in production.

None of these were urgent. All of them are the kind of thing that signals "we don't pay attention" if left untouched — so we touched them.

ChatGPT, Gemini, and Perplexity each have web-search tool access. They use it inconsistently. Sometimes the disagreement we're surfacing isn't a training-data difference — it's that one model searched live and got a fresh answer while another didn't.

The consensus matrix now annotates each answer with a small mono badge: web ↗ if the model called search, blank if it answered from training. This changes how you read disagreement: a "training-vs-search" split is different from a "the models disagree about reality" split.

Subscribers can now embed their live score as a badge — Monitor tier and above. The endpoint is cached at the CDN edge with a 1-hour TTL, so it costs us nothing per request and stays current without polling.

• 3 brands — $899/mo (Monitor) or $1,399/mo (Pro). ~25% per-brand savings.
• 10 brands — $1,999/mo or $2,999/mo. ~50% per-brand savings.
• 20+ brands — custom, contact sales until the demand pattern stabilizes.

Each bundle is a single Stripe subscription. Brand assignment is managed in-app, not in Stripe — so swapping a brand mid-cycle doesn't require a billing change.

Drift can go either direction: the AI gets it wrong, or the site changed and the AI is still right about the old version. We now distinguish those cases by re-crawling on the same schedule as the model queries. The $199 report uses a 14-day freshness window; Monitor and Pro keep a rolling baseline.

Symptom: scan starts, three model calls succeed, the batch sits at "Scoring consensus" forever. Root cause: the consensus aggregation query took longer than the pool's idle timeout. The connection dropped, the query silently returned null, and the workflow had no fallback.

Now: explicit query timeouts, structured error logging, automatic retry on the workflow step, and the batch shows "Failed — retrying" instead of pretending to work.

The first version of misquoted asked a simple question: when three different AI models answer the same question about your brand, do they agree? It turned out the answer was usually "no" — and that the disagreement was where the interesting story lived.

Everything since has been about turning that one observation into a product people will pay for. Some of those bets worked, some didn't, and this changelog tries to be honest about both.