New

• Site audits now check who can crawl your site and how it's indexed — The comprehensive site audit now inspects your robots.txt, sitemap.xml, hreflang tags, and an optional llms.txt, and reports whether each major AI crawler (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and more) is allowed or blocked. A new International & Crawlability tab shows a per-bot allow/block panel, your sitemap and robots.txt status, hreflang health, and llms.txt presence. A blocked bot is flagged as high priority; a robots.txt that blocks everything site-wide is critical. Missing hreflang or sitemap shows as "not applicable" rather than a false error. The AI Visibility dashboard gets a matching AI-crawler summary card, and the Strategy Hub factors crawlability into its technical-health and AI-visibility analysis.
• Choose the country and language for your keyword data — Keyword search volume, difficulty, rankings, and AI keyword data used to be locked to the United States and English for everyone. You can now pick a target market from the Keywords page: a market button in the header shows your active market and opens a picker with dozens of countries and languages. New keyword scans, the scheduled comprehensive scan, and on-demand research all use your selection. Existing accounts stay on United States / English unless you change it.

Improved

• A smarter 0–100 AI Visibility score — The headline score used to weigh only how often your brand was mentioned plus a crude "appears in the first 500 characters" flag. It is now a weighted metric blending four signals: how often you're mentioned (40%), how early in the answer you appear (20%, measured as a smooth position rank instead of a yes/no cutoff), how positively the answer talks about you (20%), and whether the answer cites supporting sources (20%). Older scans keep their original scores so your historical trend isn't rewritten.
• Trends flag when the underlying AI model changes — Each scan now records which AI model produced each result, and the trend chart marks any scan where the model version changed, warning that scores around that point may not be directly comparable. A shift caused by a vendor model update is no longer mistaken for a real change in your visibility.
• More accurate sentiment and brand-mention detection — The scanner now judges whether an AI answer speaks positively, negatively, or neutrally about your brand using real language understanding, correctly handling negation and competitor comparisons like "X isn't as good as Y", instead of counting positive and negative keywords. Brand-mention detection now matches whole words and recognizes your aliases, domain names, and product names, so it no longer miscounts your brand inside unrelated words (for example "top" inside "laptop") and no longer misses mentions that only reference your domain.
• Data Chat no longer gives confidently wrong "you have nothing" answers — When your website URL isn't set, Data Chat used to quietly run backlink, keyword, site-health, and competitor questions against a blank domain and report things like "you have no backlinks". It now recognizes when your site isn't connected and tells you what to add. Individual tool failures are surfaced rather than swallowed, so it explains what went wrong instead of treating an error as "no data".
• Proactive Insights nudge you instead of going silent — When there is nothing to analyze, the What's new area and notification bell used to just stay empty. They now show an actionable prompt to connect a data source so you know why there are no insights. The nudge is rate-limited so it won't nag you daily.
• Smarter article extraction when importing a blog by URL — The Blog Voice Profile URL importer now uses a reader-grade article extractor to pull the main post text out of messy, heavily styled blog pages, stripping navigation, sidebars, ads, and comment threads far more reliably. If we fetch a page that clearly has article text but still can't isolate it cleanly, you now get a "try copying the post text and pasting it instead" message rather than a flat "page was empty" error.