See what LLMs are saying about your brand.

A closed-beta SaaS for tracking and improving brand visibility in generative engines. Three things, in priority order:

1. Inaccuracy detection. Diff what LLMs say about you against your canonical facts. The flagship surface.
2. Multi-engine tracking. Run a prompt set across 5 engines (API + real browser) on a recurring schedule. Log every citation, mention, and sentiment.
3. Technical GEO diagnostics. Audit your site for the bot-accessibility, schema, and chunk-density issues that keep LLMs from picking you up.

Brand-agnostic - everything is configured per-tenant in the app. No-code outside the brand setup form.

1. Get approved. Submit the form at /request-access. The admin will email a magic-link invite once approved.
2. Sign in. Click the magic link, or use Google. You can set a backup password from your account screen.
3. Add your brand. Brands → New brand. Fill in name, domain, category, and the four list fields (aliases, competitors, personas, agent instructions).
4. Add canonical facts. On the brand page, open the Canonical facts tab and add 8-15 facts across the 5 categories (see below). This is the ground-truth for the Inaccuracy Engine.
5. (Optional) BYOK. Settings → add your own Anthropic / OpenAI / Perplexity / Gemini keys. Keys are encrypted at rest with AES-256-GCM. Skip this and tracker runs use the platform pool.
6. Run. (Available in the next release) trigger a tracker run from the brand page or wait for the weekly schedule.

Brands → New brand. Each field shapes downstream tracking - hover the (i) tooltip on the form for full guidance. Quick guide:

• Brand name - the canonical capitalization. Used as ground-truth for spelling/casing checks.
• Domain - apex domain, no protocol. The audit suite crawls from here.
• Category - specific market category. Drives prompt expansion.
• One-liner - your positioning statement. Lands in your generated llms.txt.
• Brand aliases - all the variants. Misspellings, abbreviations, casing, parent-company name, the bare domain. More aliases → fewer false negatives.
• Competitors - tracked alongside you. Include obvious AND long-tail.
• Target personas - 3-5 specific job titles. Drives prompt expansion.
• Agent instructions - one per line. Embedded in your generated llms.txt.

The single most valuable input you can give LLMSVIZ. Each fact is one statement an LLM should be able to make about your brand without getting it wrong. The Inaccuracy Engine (coming soon) uses these to flag every wrong claim it finds in tracker runs.

LLMSVIZ supports two modes for the LLM calls a tracker run makes:

• Platform pool - default. We use our shared keys, no setup needed. Free during closed beta.
• BYOK (bring-your-own-keys) - paste your Anthropic / OpenAI / Perplexity / Gemini keys at Settings. Encrypted with AES-256-GCM at rest. We display only the last four characters; the plaintext is decrypted only at run time. Useful if you want isolated billing or higher rate limits.

Each engine has its own quirks. We track all five and surface engine-specific recommendations:

• ChatGPT - mirrors Bing top-10 about 87% of the time. Setting up Bing Webmaster Tools is the highest-ROI move for ChatGPT visibility.
• Claude - heavier reliance on parametric memory and curated training data. Brand recognition is a longer game; community signals and docs presence dominate.
• Perplexity - around 47% of citations come from Reddit. Real-time retrieval; fastest engine to reflect new content.
• Gemini - pulls from Google's organic index. About 76% of cited URLs rank in Google top-10. Traditional SEO is a prerequisite.
• Google AI Overviews - the answer box on google.com. Distinct from Gemini API. Reduces top-organic CTR by ~59% when present.

Two test modes per engine, run on every tracker pass:

• api - direct API call. Tests parametric memory when web search is disabled, retrieval when enabled.
• browser - real Playwright session against the public web UI. Captures what your customers actually see.

One-click crawl of your domain that scores GEO readiness across the dimensions LLMs actually care about:

• robots.txt rules per-bot (GPTBot, ClaudeBot, PerplexityBot, Googlebot-AI, OAI-SearchBot, Bingbot)
• SSR vs CSR - do AI crawlers actually see your content?
• Schema completeness (Organization, Product, FAQ, Article)
• llms.txt presence and validity
• ai-plugin.json (ACP / agentic readiness)
• AI Overview presence on your top keywords
• RAG chunk simulation - is your page chunkable?
• Answer Capsule check - are 40-60 word direct answers up top?
• FCP via Lighthouse

Each check returns pass/fail/warn with a remediation note and severity ranking.

From your brand config and canonical facts, LLMSVIZ generates the assets agents need to find and represent you correctly:

• llms.txt - Stripe-style instructions to AI agents
• JSON-LD schema blocks (Organization, Product, FAQ)
• ai-plugin.json for Agentic Commerce Protocol readiness
• Information Gain scorer - rates a page on five dimensions and suggests improvements
• Prompt expander - takes your seed prompts and generates 30-100 realistic variations with persona biasing

Outputs are downloadable. We never auto-publish to your site.

What ships when. Live = available now, Coming soon = next release, Planned = on the path.

• Live Closed-beta access flow with admin approval queue
• Live Brand configuration with aliases, competitors, personas, agent instructions
• Live Canonical facts editor across 5 categories
• Live BYOK API keys with AES-256-GCM encryption
• Coming soon 5-engine tracker (ChatGPT, Claude, Perplexity, Gemini, AI Overviews) - API mode
• Coming soon Real-browser tracking via Browserbase
• Coming soon Inaccuracy Engine - the flagship surface
• Planned Technical audit suite
• Planned Generators (llms.txt, schema, ai-plugin.json) and Information Gain scorer
• Planned Competitive Landscape view + Crawler Budget tracking + weekly digest

• We never auto-publish content to your site. Generators output files; you deploy.
• We don't scrape Reddit, HN, or community platforms to influence sentiment. Astroturfing is a model-blacklist risk and we won't help with it.
• BYOK keys are stored encrypted at rest with AES-256-GCM. The master key lives in our platform secret manager and never touches the database. Plaintext keys are decrypted only at tracker-run time and never returned to the browser.
• Every key save, replace, delete, and decrypt event is logged to an append-only audit trail.
• Tracker runs are append-only. We don't edit historical measurements; if a re-classification is needed, we add a new row.