# AgentLens

> AgentLens measures how AI assistants discover, read and recommend your brand across ChatGPT, Claude, Gemini, Grok and Google AI Mode, then turns the evidence into fixes you approve and a re-measurement that proves the change.

AgentLens is the Agentic Visibility Platform (Beta). It measures AI visibility and discoverability for agentic commerce, generates reviewable remediation and commerce artifacts, and re-measures what changed. It never silently modifies a customer website.

## Pages

- [Home](https://agent-lens.ai/): the vision, the shift to agent-mediated discovery, the scorecard and the operating loop
- [Platform](https://agent-lens.ai/platform): every capability organised by stage: configure, measure, diagnose, improve, integrate, govern
- [How it works](https://agent-lens.ai/how-it-works): three ways in, the six-stage loop, pillar formulas, the estimate-confirm-run spend contract
- [Trust](https://agent-lens.ai/trust): what AgentLens does not do, controls, platform status, FAQ
- [Sector reports](https://agent-lens.ai/sector-reports): the most visible brands of each published sector, read from one shared set of prompts
- [Roadmap](https://agent-lens.ai/roadmap): Live / Now / Next / Later from the roadmap catalog
- [Request a report](https://agent-lens.ai/request-report): form to request an agentic visibility report (domain + at least three prompts)
- [Privacy and KVKK notice](https://agent-lens.ai/privacy)
- [Terms](https://agent-lens.ai/terms)

## Measured AI surfaces

- ChatGPT (OpenAI): GPT models via the OpenAI API
- Claude (Anthropic): Claude models via the Anthropic API
- Gemini (Google): Gemini models via the Google API
- Grok (xAI): Grok models via the xAI API
- Google AI Mode (Google Search): Google's own AI answer for the prompt, read as a live search surface

Perplexity is not measured because no real adapter exists for it.

## Scorecard

- Schema Coverage (25%): JSON-LD structured data: presence, correctness, completeness. Formula: (pages with valid JSON-LD ÷ pages crawled) × 100
- Content Structure (25%): Semantic HTML, meta quality, heading hierarchy, llms.txt. Formula: semantic HTML (25) + structure (25) + meta (25) + llms.txt (25)
- Agentic Brand Familiarness (25%): How often AI mentions your brand organically: Share of Model. Formula: (mentions found ÷ prompts tested) × 100
- Shopping Assistant (25%): Position and frequency in AI product-recommendation shortlists. Formula: Σ (points per query × position multiplier)

## Operating loop

1. Configure: Brand, prompts, rival pool, AI surfaces and cadence live in Brand Management. Saving creates a new prompt-set version; history is never rewritten.
2. Measure: A spend-free estimate shows pinned models, expected calls and a hard cap. Only after you confirm does a run ask the real vendor models.
3. Diagnose: Four-pillar score with confidence intervals, an evidence map from every score back to its page or answer, prioritised issues and guides.
4. Improve: Fix Engine and Content Optimizer generate deployable artifacts from the run: JSON-LD, llms.txt, robots.txt, meta tags, rewritten content, feeds.
5. Approve & publish: You review; your own pipeline deploys. The treatment is recorded as an intervention so the next measurement can be compared against it.
6. Re-measure: Scheduled cycles report what changed and whether it beat sampling noise. Alerts reach high severity only when a significance test rejects noise.

## Capabilities

### Configure
- System Prompts: Measure the governed prompt set saved for your sector and brand: Turkish banking ships 168 prompts across retail, SME and corporate segments.
- Custom Prompts: Ask ad-hoc questions without touching the saved taxonomy. The right way to test a campaign or a product launch.
- Data Feeding: Audit a product feed (CSV, JSON, XML) with no crawling at all, for sites that block crawlers or catalogs that have not shipped yet.
- Brand Management: Version your prompts, sector rivals, AI surfaces and cadence. Saving creates a new prompt-set version and never rewrites history.
- Prompt taxonomy: Recommendation, research and comparison intents, scoped by customer type, product family and channel, so results can be sliced later.

### Measure
- Five surfaces: ChatGPT, Claude, Gemini and Grok through their vendor APIs, plus Google AI Mode read as a live search surface.
- Estimate → confirm → run: A spend-free estimate with pinned models, expected calls and a hard cap at 115%. A run exists only with a current, single-use estimate and your confirmation.
- Smallest detectable change: The estimate states the least week-over-week movement the run could confirm, so you never buy a benchmark that cannot answer the question.
- Variance control: Multi-trial averaging, pinned model versions, deterministic sampling and a 95% confidence interval on every metric.
- Bring your own keys: Save, test, rotate and delete OpenAI, Anthropic, Gemini and xAI keys. Encrypted at rest; owners and admins only.

### Diagnose
- Smart crawler: Sitemap discovery, two-stage sampling, a 7-tier fetch cascade for WAF-protected sites and optional headless rendering for JS-heavy pages.
- Schema & content audit: Per-page JSON-LD extraction and validation, semantic HTML scoring, meta quality and llms.txt detection.
- AI bot access check: Detects whether GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot and Googlebot are blocked in robots.txt.
- Evidence map: A full causality trail from every score back to the page, tag or model answer that produced it.
- History & diff: Audit history, snapshot comparison, added and removed page detection, per-audit token and cost tracking.
- Action queue: Ranked, evidence-backed work items assembled from what the pipeline recorded, not a generic checklist.

### Improve
- JSON-LD generator: Complete structured-data blocks for the pages that lack them.
- llms.txt generator: A spec-compliant file for your root, built from brand identity and the crawl manifest.
- robots.txt fixer: Unblocks the AI crawlers you want and keeps the ones you do not.
- Meta tag optimiser: Title, description and Open Graph tags rewritten for the page.
- Semantic HTML audit: Pattern detection plus fix snippets for the markup agents read.
- Fact-Feel-Proof rewriter: Three-block product descriptions written for model comprehension.
- Doc2Query++: Synthetic queries plus FAQPage JSON-LD, so the page answers the questions people ask assistants.
- Image audit & attributes: Missing-alt detection with AI captioning, and inferred attributes enriched into your JSON-LD.

### Integrate
- ACP · OpenAI: ACP-compliant JSON and TSV product feeds with validation.
- UCP · Google: Merchant Center JSON and XML with conversational attributes.
- MCP · Anthropic: A Docker-ready Python MCP server, scaffolded and downloadable.
- Relationship tagger: Substitute, accessory and cross-brand relationships across your catalog.
- Agent-first API: An OpenAPI spec plus a FastAPI sample you host, so agents can query your catalog directly.

### Govern
- Automations: Scheduled re-measurement delivered to Slack, Microsoft Teams, email or a webhook, with an alert inbox. Each cycle runs under its own hard cap; one that would exceed it is refused.
- Alerts that survive a significance test: A week-over-week drop reaches high severity only when a two-proportion test rejects noise at α 0.05. A real drop names its metric in the subject line.
- Sources to earn: Cited sources that informed answers naming a rival but not you: a ranked, exportable list of places to get listed.
- Benchmark exploration: Slice visibility and comparison results by scope, topic, intent, provider and active rival pool. Dashboard and report filters agree.
- Exposure ledger & snapshots: Portfolio trends, per-brand breakdown, weakest and strongest pillar; immutable report snapshots as PDF, PPTX, XLSX and JSON.
- Sector catalogs: Turkish banking (47 banks, tiered rival pools) and fashion (99 tagged brands) today. Fintech, tech & electronics and lodging as approved packs.

## Sectors

- Banking & Finance: 47 banks, tiered rival pools
- Fashion & Apparel: 99 tagged brands
- Fintech & Payments: approved pack
- Tech & Electronics: approved pack
- Lodging & Travel: approved pack

## Published sector reports

Visibility is the share of measured AI answers that name the brand, across one shared set of prompts per sector.

- Banking & Finance: Banking & Finance · 47 brands · v1 (published 2026-09-30). Most visible: İş Bankası 62.5%, Garanti BBVA 60.8%, Akbank 52.4%, Yapı Kredi 50.7%, Ziraat Bankası 42.7%.
- Fashion & Apparel (sub-category Style: sporty only): Fashion & Apparel · 13 brands · v1 (published 2026-09-28). Most visible: Nike TR 75.6%, Adidas TR 74.1%, Decathlon TR 69.6%, Puma TR 51.8%, New Balance TR 44.1%.

## Boundaries

- No silent changes to your site: AgentLens never deploys to your CMS, storefront, robots.txt, DNS or merchant account. It generates; you apply.
- No guaranteed rankings or sales lift: Scores and projections are decision support. Model behaviour changes; we show confidence intervals and significance, not promises.
- No transactions: No checkout, no payment, no autonomous purchasing on your behalf.
- No surfaces we cannot measure: If there is no real adapter for an AI surface, it is not offered or advertised. Perplexity is not on the list today for exactly that reason.

## FAQ

Q: Which AI surfaces does AgentLens measure?
A: ChatGPT, Claude, Gemini and Grok through their vendor APIs, and Google AI Mode read as a live search surface. Perplexity is not measured because no real adapter exists for it.

Q: Does AgentLens change my website?
A: No. It generates deployable artifacts such as JSON-LD, llms.txt, robots.txt and meta tags. You review them and your own pipeline deploys them.

Q: How is the readiness score calculated?
A: Four pillars at 25% each: Schema Coverage, Content Structure, Agentic Brand Familiarness (Share of Model) and Shopping Assistant. Each is scored 0–100 with a 95% confidence interval and its sample size.

Q: Will I know the cost before a run?
A: Yes. Every run starts with a spend-free estimate that shows pinned models, expected calls and a hard cap at 115% of the estimate. Nothing runs until you confirm.

Q: Is AgentLens generally available?
A: It is in beta, verified in a shared staging environment. Beta pilots run there; no production environment is claimed.

## Contact

- Request a report: https://agent-lens.ai/request-report
- Email: contact@agent-lens.ai
- WhatsApp: +90 539 459 17 06
- LinkedIn: https://www.linkedin.com/company/agentlens-ai/
- Address: İstanbul, Türkiye
- Platform sign-in: https://platform.agent-lens.ai