What we've shipped, newest first. Tagged New, Improved, Fixed, and Honesty (corrections that removed misleading copy, counts, or charts).
Subscribe by RSSThe public homepage check now describes its three supported checks: title, meta description, and first H1. Workspace setup explains unavailable competitor suggestions and keeps manual entry available. Accounts without a workspace return to setup, and unavailable password recovery is shown explicitly. Payment-reminder handling preserves uncertain delivery states instead of reporting them as delivered.
Billing now prevents overlapping checkout actions and ignores responses for an earlier workspace or selection. Plan access and API usage settlement use current authority, with unconfirmed settlement disclosed. The calculator keeps usage estimates separate from account balances and checkout amounts; prices, allowances, and the policy against automatic overage charges are unchanged.
Every public page was audited claim-by-claim. Corrected: language count (the true content-language registry count), uptime/SOC 2/GDPR badges now match the trust page, a retired Google AI-product brand name replaced with the real engine registry incl. You.com and Phind, the usage-stat strip replaced with capability stats, and the status page no longer reads "all systems operational" over an empty component registry. The 2026-06-24 experiments entry was amended: comparison results are directional — exposure collection is not yet a controlled experiment system, so no "winner" claims. BYOK left public pricing until real scheduled per-workspace routing ships. The forbidden-claims list now lives in code with a contract test over every marketing surface and all 18 locales.
The methodology page now leads with what actually ships: Wilson confidence intervals over effective sample sizes, a calibration backtest on the intervals themselves, a separate branded-prompt lane, and a schema-drift canary — each citing its source file. New public pages: /capture (per-engine capture-mode matrix, generated from the engine registry), /facts (canonical code-derived fact sheet), /methodology/reading-ai-visibility-claims (a six-question buyer checklist), /compare (dated, sourced comparisons that name where competitors win), and /bot (our crawler user agents and how to block them). Plus robots.txt, sitemap.xml, structured data, and an RSS feed for this changelog.
User-supplied branding in white-label PDF/HTML reports is now escaped before render, closing a stored cross-site-scripting hole in agency report exports.
The sidebar now collapses to a compact icon rail from a single toggle, and the choice persists across pages and reloads. Studio opens collapsed so the writing canvas gets the space; every other surface remembers your preference.
The Studio editor now matches the rest of the workspace — consistent surfaces, selection toolbar, and inline AI-suggestion styling. The underlying TipTap engine and all scoring is unchanged; only the look was updated.
The small diagonal line beside the Studio content score implied a tracked trend, but it only reflected score changes within the current editing session and reset on reload. It is gone, replaced by an honest note that the score is a live weighted blend of the six factors below it.
Technical gained a Decay tab (a measured retroactive scan plus a labelled forecast model) and a Localization tab (hreflang reciprocity checks, a cited-in-English-but-absent-in-language gap detector, and a regulatory lint list). Both are per-workspace with clear setup and no-data states.
The consolidated workspace now includes the five core differentiators, four GEO visibility lenses, A/B experiments, and roughly twenty surfaces and tabs—each backed by a real engine or a clearly labelled empty state.
The primary navigation is now banded into Monitor, Create, Research, and Optimize so a growing surface count stays scannable instead of becoming a long flat list.
The Stability lens reports citation probability with a Wilson confidence interval, and stays honestly inconclusive when there are too few repeat observations per engine to compute one — no fabricated reproducibility numbers.
Rankings now flags pages whose organic position holds while AI citations fall, and Site Explorer ledgers which crawled pages are not yet cited — the fastest GEO wins — with citation status shown as a true tri-state, never a false "not cited".
A dedicated Experiments surface runs title, meta, and CTA comparisons with z-test significance labeling. Comparison results read as directional evidence: exposure collection is not yet a controlled experiment system, so the surface does not declare winners.
Technical gained an Algo watch tab combining Google update signals with AI-answer volatility into one weather report, with per-workspace impact derived from your own data rather than a global feed.
The Technical navigation item carried a hardcoded "4" badge that did not reflect any real finding count. It was removed — counts now only appear when derived from actual findings.