What's changing at GeoStars
One entry per real change, newest first. No catch-all items, no internal ticket numbers – and if nothing happens in a given week, nothing appears here either.
Ads Radar
Shows sponsored placements in the AI answers to your prompts: who is advertising, with which page, and on which model. Every new answer is checked automatically.
7-day overview by email and Slack
If you like, you can get the weekly overview from "Analyses" every Monday by email or in your Slack channel.
Trial with payment method
You now start the 7-day trial with a payment method on file. You're only charged after 7 days, and you can cancel anytime with one click before that – and we'll remind you by email before the trial ends.
No ad tracking in the app
Google ad tracking no longer runs in the logged-in app. Only the sign-up page measures anything – and even that only with your consent.
New metric tiles
All metrics now follow the same clear pattern: a colored icon, a large number, a short explanation underneath. The numbers are shown neutrally in black.
Topic prioritization with table
Next to the matrix there's now a table with visibility, share of the prompt set, and quadrant per topic – so even closely clustered topics can be read unambiguously.
Top 10 sources, clearer
The most-used domains are shown in two columns with rank, bar, and your own domain highlighted.
Explanations in the prompt list
Web share, intent, search volume, and Difficulty each now have an explanation window: what the column measures and how to read it.
Ownership and due date on the board
On every recommendation card you can now set who handles it and by when.
Outage notice based on real measurements
The notice for a disrupted platform now appears when no measurement has succeeded there for more than 26 hours, and disappears automatically once the first one succeeds.
Google AI Mode reported separately
In the prompt detail, Google AI Mode now has its own row. Previously it was counted within Google AI Overview.
Bug fixed: trend vs. change
The trend indicator and the change versus the previous period sometimes pointed in opposite directions for the same brand. Both now calculate on the same basis.
Bug fixed: model switcher in prompts
After switching to "All models," the numbers for the core models sometimes stayed on screen briefly after loading. The page now always shows the selected choice.
Fully in English
Prompt detail and Perception are now fully translated; dates and percentages follow the selected language.
Google AI Mode as standard
New prompts now automatically measure all six platforms – Google AI Mode runs every three days, just like Perplexity and Microsoft Copilot. Selectable in every package.
Usage at a glance
The usage page and header show the credits actually used – the same number billing is based on. What your setup is projected to use over the month is shown below as a forecast.
Trends: risers and fallers
A new page shows which brands, prompts, and sources have improved or worsened over 7, 14, 30, and 90 days – side by side and with export.
Metrics explained
Every key metric has a “More about this” panel: what it measures, how it's calculated, how to read it, and what you can do about it.
Export with filters
AI metrics, prompts, and sources can be exported as CSV. The file names all the filters that were active during export.
Prompt types
Prompts are now grouped into organic, competitive comparison, and brand-related, and can be filtered accordingly.
Empty fields stay empty
Where no measurement exists yet, the dashboard, reports, and report templates show a dash instead of 0%. A report without measurement data consists of a notice slide instead of zeros.
A tidier dashboard
The chart and metrics fit on one screen without scrolling; notes appear where they're needed.
API: typos are reported
The API responds to unknown parameter names with error 400 and the list of allowed names. from and to work as documented, and the limit is 120 requests per minute and key.
Write permissions on the MCP server
Deleting and activating via the MCP server now require a key with write permission. Previously, some of these actions also worked with a read-only key.
Numbers in your language
Numbers and dates now follow the selected language everywhere – in English it shows 7.8 instead of 7,8. This applies to metrics, tables, charts and recommendation texts.
Monitors by target audience
You can split prompts into monitors, for example by target audience or country, and evaluate each monitor separately. Billing stays the same.
Status page with causes
Every outage is labeled as a provider outage, a cross-platform outage, or our own outage; the most recent outage is shown at the top.
Difficulty per prompt
The prompt list shows how hard it is to appear in the answers (0–100). The formula is shown openly in the "Learn more" window.
Prompt Explorer
Find new prompts from your Search Console and from in-app suggestions – with source, impressions, and one-click adoption.
Recommendations as a board
Recommendations can now be moved through Suggested, To Do, In Progress, Done, and Discarded like on a board – or viewed as a list.
Page Tracker
For every page of your own, you can see how often AI crawlers fetch it, in how many answers it's cited, and how many visits result from it.
Analyses at the push of a button
Four ready-made reports: 7-day overview, visibility diagnosis, prompt setup check, and Search Console opportunities – every number drawn from your measurement data, copyable as text.
Content calendar
You can now also see planned and published articles as a monthly calendar.
Visibility you can verify
For every visibility figure, we show the numerator and denominator: how many answers there were and in how many your brand appears. Citation rate and crawl rate are explained separately.
Prompt list now counts everything
For prompts with very many answers, the prompt list only counted part of them. Fixed – we removed the same limit in three other places.
Clearer team permissions
Only owners and admins can now change project settings. Every project member can see their project's member list.
More accurate brand recognition
A second verification step corrects clear misattributions involving your brand and competitors before they enter the metrics.
API filters now work reliably
The model and country filters now apply across all v1 endpoints. The API rejects unknown values with a 400 error instead of silently ignoring them.
Google AI Mode
Google AI Mode is now tracked too – as the sixth platform in the extended class, every three days by default. It only moves into the daily core measurement after two weeks of proven stability.
A finer sentiment scale
Sentiment is now scored per brand on a scale from 0 to 100; a neutral mention in a list sits around 50. Existing values were recalculated once.
A trial instead of a free-forever plan
New accounts start a 7-day trial with up to 50 prompts. After that, GeoStars continues measuring with a booked package.
Email settings
You decide which emails you receive. Every optional email includes an unsubscribe link; marketing emails and FactCheck alerts are switched off by default.
A tidier navigation
The sidebar now has eight sections, with subpages shown as tabs above them. All previous addresses continue to work.
AI traffic on your website
A cookieless counting script (under 2 KB) shows how many visits come from ChatGPT, Perplexity and similar – by platform, day and landing page. We deliberately don't report Google AI Overview separately: those visits can't be cleanly separated from normal Google search.
The knowledge base fills itself
When you create a project, GeoStars reads your website and builds a fact profile with source links. FactCheck automatically checks AI answers against it.
From source gap to article
A gap in the source analysis becomes a briefing, and the briefing becomes a draft. After publication, GeoStars measures whether the page gets cited.
Fanout searches
The search queries that AI systems trigger in the background now have their own page – with text search and CSV export.
Bot visits and crawlability over time
The bot analysis shows the last 24 hours hour by hour and links the first bot visit to the first citation for each URL. The crawlability check saves a snapshot weekly and shows which bots have been newly blocked or allowed again.
System status, API docs and llms.txt generator now public
Three pages without login: measurement success per platform over 30 days, the documentation for the REST API and MCP server, and a free llms.txt generator.
MCP with OAuth login
Claude on the web and in Cowork connects to GeoStars via login. You no longer need to copy an API key for it.
Google rankings per prompt
For every prompt keyword, GeoStars measures the classic Google position – including SERP elements like AI Overview, Featured Snippet and “People also ask”. Competitors in the top 20 are flagged.
Brands matched as whole words only
Brand names are now only recognised as whole words. A brand called “Star” no longer counts as a match inside “Starbucks”.
No more Google fonts in the app
All fonts are now hosted on our own servers – the app no longer sends a request to Google when loading. Terms and Conditions and the privacy policy are now also available in English.
This log starts on 12.09.2026; we don't add older changes retroactively. How reliably measurements are currently running is shown by the System status.
Set your course
today.
Start free – in a few minutes you will know where your brand stands across the AI universe.