Ansvisor Files

Added

• Google Analytics 4 integration — connect a GA4 account from Settings → Integrations through Composio managed auth, map each brand to one of its properties, and sync AI-sourced sessions, conversions and revenue daily (#695, [#703], [#718]). AI Traffic then reports what AI referrals are actually worth, not just how many arrived, and instances without Composio credentials degrade to a safe no-op
• Prompts can target several locations at once — a prompt's tracking locations are now a per-prompt list of countries and US states (US-CA), edited from the prompt card with a searchable picker (#783). Plan quota meters the tracked location rather than the prompt row, since a prompt tracked in three places costs three times the scraper and model calls; every write path measures the change it makes, so re-saving at the cap succeeds and removing a location frees capacity immediately (#781). Google AI Overview and AI Mode have no sub-country mechanism, so their state locations collapse to one country-wide run instead of billing the identical query twice
• Per-prompt visibility broken down by location — a Location control on the Prompts page re-reads the health columns for one tracked location, so a prompt that is strong at home and invisible abroad no longer hides behind a blended average (#784)
• Pages AI engines ignore — analytics-backed detection of valuable pages that AI search sends no traffic to, surfaced in AI Traffic and turned into prompt suggestions built from the pages worth defending (#724, [#725], [#720])
• Prompts: suggestions get their own tab that shows what feeds them (#727), a new prompt is analyzed the moment it is added (#754), and a 7d / 30d / 90d range selector (#697)
• One date range control across Visibility, Citations and Prompts, so the three pages can no longer disagree about the window they describe (#722)
• Tracking: an OpenRouter provider, with sentiment analysis made provider-agnostic (#708)
• Reports moved into the sidebar (#750)
• Ops: CI flags pull requests that change a migration, so the apply step is not forgotten (#736)

Changed

• Citations, Topics and Insights aggregate in Postgres instead of downloading their windows. The overview pages stopped shipping tens of thousands of rows to the browser to add them up (#726, [#744]); the per-URL detail page and Competitor Gaps read the citation index instead of paging every answer, which also removed a silent 50,000-row truncation that had been hiding the newest answers from both (#776, [#778]); and the dashboard serves from daily rollups rather than rescanning history (#773). Citations are written as rows alongside each answer to make this possible (#737, [#739])
• Content Optimization generates three opportunities per run instead of five to fifteen, so a run produces work a team can actually act on (#735)
• Prompts: the All Prompts table sorts by visibility score descending by default (#714)
• Citations: source filtering became a segmented control beside the table it scopes, replacing the server-side source filters (#742, [#748])
• Relative-time formatting is resolved through one shared helper instead of per-page copies (#767, [#772])

Fixed

• Security: the citation RPCs enforce organization membership. Six security-definer functions could be called with any brand UUID by any authenticated user, leaving the tenant boundary to the web tier alone; they now check membership themselves (#779, [#780])
• Settings: organization updates are restricted to admins, with server-managed columns locked against client writes (#755)
• A brand's first day no longer sends two Daily Pulse emails — the catch-up sweep treated the deliberately unpulsed onboarding run as a missed pulse (#774), and a pulse can no longer be sent twice for the same tracking window (#717)
• Tracking runs survive slow and stalled Cloro queues — runs whose first result lands late are no longer discarded (#710), ghost tasks no longer hold a run open (#690) or make a watched run sit through the full thresholds (#761), the pending-task read is paged so large runs aren't cut short (#716), and Grok submissions stop while the provider returns 500 for all of them (#734)
• Citations: answers with many citations are no longer dropped by unchunked URL lookups (#738), the stored lookup pages past PostgREST's 1000-row cap (#740), URLs resolve through the request body instead of an over-long query string (#741), "has any citations" is answered without counting every row (#707), content stays visible during refetch (#775), an ambiguous column reference is resolved (#743), and a load failure no longer toasts after leaving the page (#696)
• Prompt volumes: a brand's keywords are analysed when its setup finishes rather than only after checkout (#765), volume analysis covers a whole brand instead of failing past 50 prompts (#747), DataForSEO is asked once per brand instead of once per prompt (#749), and progress is reported from the run and picked up again on reload (#751, [#752])
• Reports: sections aggregate in Postgres and a report survives one failing section (#762); query fan-out ranks over the whole window and shows per-engine coverage (#764)
• Content Optimization: filters stay reachable on zero results and selections can be deleted (#733); metrics are calculated across all opportunities (#700); the empty-state Generate button respects the cloud restriction (#699)
• Topic detail trend charts and reports use the AI Visibility Score formula (#685, [#693])
• Slugs transliterate non-ASCII letters instead of deleting them (#763)
• CSV: fields containing lone carriage returns are quoted (#758)
• Brand avatars fall back instead of rendering nothing — a manually-set logo URL, then the primary domain's Google favicon, then the site's own /favicon.ico, then initials. Google answers an unindexed domain with a 16×16 grey-globe placeholder rather than a failure the browser reports, so it is detected by size instead of waiting for an error that never fires (#759, [#782]). Brand Settings gains a Logo URL field for the manual override, and a derived favicon is no longer written to logo_url at brand creation or on a domain change — the column now means "the icon this brand chose", which is what lets the chain run at all
• Mobile navigation preferences are aligned with the sidebar (#771)
• Database: the missing cloro_pending_tasks migration was added (#692) and the prompt_suggestions table definition recovered (#711)
• CI: messages/*.json is covered by the format check (#698)

Docs

• The README is retitled and rewritten around AI Search intelligence (#769, [#770])
• Migration 00055's comment is corrected — there was no schema drift (#723)

Contributors

Huge thanks to everyone who shipped code in this release — and a special welcome to our first-time contributors: @mikam44700, @shahzainshafique, @dawNotPoi, @luozejian and @kn-dev-code 🎉

Thanks as well to our returning contributors @BharadwajKanneveti, @Maqbool61 and @Pavel-glitch-ui for coming back with more! 🙌

Full changelog: https://github.com/ansvisor/ansvisor/compare/v0.2.0...v0.3.0