Native Web Search, Current Models & Prompt Tracking Rebuilt

Features

• Native Web Search with Real Citations — Prompts sent to xAI, Perplexity, and Gemini now run through each provider's own search, on that provider's own interface, instead of a generic search tool bolted onto a chat call. The sources recorded against an answer are the structured citations the provider actually returned, kept alongside the links parsed out of the answer text, so a citation is now evidence rather than an inference about what the model probably read.
• Member Invites — Invite people to your organization by email address. They receive an invitation, land on a dedicated acceptance page, and join without an administrator provisioning anything by hand. Invitations that expire, that were already claimed, or that belong to a different email now say exactly that instead of failing silently, and acceptance works whether or not you were signed in when you opened the link.
• Guided Setup — New accounts open on a step-by-step setup walkthrough that carries you from a bare workspace to first data, rather than an empty dashboard with a list of things you might do.
• Bot-Block Disclosure — When your host blocks our crawler rather than your site simply being unreachable, the app now says so plainly on the surfaces where it matters — crawling, connecting WordPress, and publishing — and gives you a copy-ready support ticket to send your hosting provider, with the specific request they need to action.

Improvements

• Current-Generation Models — The prompt-execution picker and the Agent Journeys agent roster were both moved to current-generation models across the vendors we track, replacing entries that had been superseded or retired.
• Prompt Tracking, Rebuilt — A correctness and design pass across Prompt Tracking. Mention rate is now the true share of executions that named your brand. Per-model views include executions that ran on a provider's own interface, which were previously missing from them entirely. Each row states the model that actually answered rather than the one that was requested, bare model identifiers resolve to their real vendor, and provider naming and logos are consistent everywhere they appear. The layout was reworked around that data so the important numbers read first.
• Journey Template Categories — Template browse now opens as collapsible category bands, so a long gallery can be scanned by theme instead of scrolled through, and suggested templates are drawn from the workspace you are actually in.
• Mobile Pass — A broad pass across small screens: pages, tables, inputs, and tab strips no longer clip at the edge, touch targets meet a usable minimum, and the settings drawer traps focus properly and locks background scrolling while it is open.
• Faster Navigation — Seven pages diagnosed as laggy were moved onto cached, cancellable data fetching, so revisiting a page shows what it already has while it refreshes, and navigating away cancels the requests you no longer need. A hidden part of the interface that was re-rendering on every navigation now stays idle.
• Dashboard Surfaces — A layering pass across the logged-in pages: consistent surface elevation, fills, and edges, so panels, sheets, and inputs read as one system rather than as pages built at different times.
• Provider Timeouts — Calls to model providers are now bounded by an explicit time budget, and a provider dispatched directly owns its own envelope rather than inheriting one meant for a different path. A slow provider now fails as a slow provider instead of stalling the work around it.

Fixes

• Rank History Dates — The Share of Voice rank-history chart now plots the dates its data was actually recorded on, instead of a regularized axis that could imply a measurement on a day nothing was measured.
• Brand-Name Prompt Generation — Generated prompts are now built from your brand name rather than the root of your domain, so a brand whose name differs from its URL no longer gets questions about a word nobody would search for.
• Duplicate Competitors — Fixed duplicate competitor entries being created, and cleaned up the duplicates already present so competitor comparisons count each rival once.
• Publishing Confirmation — A confirmed publish that then fails no longer leaves the confirmation stuck as satisfied; the next attempt asks again.
• WordPress on Challenge-Protected Hosts — A host serving a bot-challenge page with no title is now recognized as a challenge rather than mistaken for a valid empty page, so crawls, journeys, and WordPress publishing report a block as a block and carry that reason through to the error you see.
• Gemini Capacity Handling — Requests rejected for capacity, and requests that time out for the same underlying reason, are now retried and fall back to a sibling model instead of surfacing as a failed execution.
• Model Row Identity — Model rows that arrived under an unprefixed or inconsistent identifier are canonicalized, so one model no longer appears as two entries with the score split between them.