# CitePulse

> CitePulse (citepulse.ai) is an AI citation audit and monitoring platform operated by GeoFlow Solutions Marcin Gamża. It measures whether AI assistants — ChatGPT, Perplexity, Gemini, Claude, Grok and Google AI Overviews — cite a company in answers to its customers' real buying questions, flags when AI states something false or out-of-date about it, and names the competitors recommended instead. It tests the models' live answers, not static page source.

Key facts:
- AI Reputation Score (0–100): built from four stacked layers — Readable (can AI crawl the site), Understood (does AI grasp the entity), Cited (does AI name the company when buyers ask — the money layer), and Quotable (source-backed material AI can lift). A weak lower layer caps the ones above it.
- Citation share: the percentage of AI answers to a fixed set of buyer questions in a category where a domain is cited as a source, relative to all cited domains, per engine, tracked over time — the share-of-voice metric for AI search.
- Engine ladder: the free mini-audit covers ChatGPT, Perplexity and Gemini (Google AI Overviews read as a source); the one-time Full Audit adds Claude; the Protect and Scale plans add Grok (xAI). Up to five assistants. Each buyer question is asked several times for stability, and detection is deterministic (brand name in answer text, domain in cited sources).
- Fact Integrity: the client approves the facts AI must get right (prices, key numbers, claims); CitePulse checks whether AI answers, the client's own site and the third-party listings AI cites still state them correctly, flagging "drift" when AI contradicts an approved fact and "stale echo" when AI repeats a retired value, with the source to fix. Included in every monitoring plan (Shield 10, Protect 20, Scale 30 approved facts).
- Agent Readiness (2026): a separate, forward-looking score for the agentic web — whether an AI agent can find, read and act on a business, not just quote it. Layer A (Legibility): server-rendered content, llms.txt, machine-readable pricing, AGENTS.md. Layer B (Commerce, storefronts only): product schema, shipping/return policies, agentic-checkout signals (e.g. a Universal Commerce Protocol profile) — never counted against a services/SaaS site. Scored separately from the AI citation score.
- Verifiable Brand Facts: approved facts are published as a signed, versioned feed agents verify cryptographically (Ed25519/EdDSA, public JWKS) without scraping — feed https://api.citepulse.ai/.well-known/brand-facts/citepulse.ai.json, keys https://api.citepulse.ai/.well-known/citepulse-jwks.json, single-claim verify https://api.citepulse.ai/api/facts/verify/citepulse.ai?claim=<key>, badge https://api.citepulse.ai/facts-badge/citepulse.ai.svg. The signature attests the brand APPROVED a value as of a version, not objective truth.
- Brand-as-MCP: each brand can get a read-only Model Context Protocol endpoint agents (Claude, Cursor, Codex) query directly — https://api.citepulse.ai/mcp/citepulse.ai (tools: get_facts, get_pricing, get_policies, get_products; responses Ed25519-signed). Discovery manifest: https://citepulse.ai/.well-known/mcp.json.
- Brand Defense: detects hidden AI-manipulation text — prompt-injection / recommendation-poisoning concealed from humans but readable by AI crawlers — and never flags a page that merely discusses these topics in visible copy.
- Agent Simulation: models the discover→read→buy journey and names the step where an AI agent would drop off (deterministic on the free audit; a live reasoning pass on paid audits also names which competitor an agent would pick).
- Auto-Fix: copy-paste files generated from an audit's gaps — Organization schema, llms.txt, robots.txt, sitemap and FAQ schema.
- Doors A/B (DOORS-1.0, since 2026-09-05): for every buying question CitePulse classifies the sources the AI engines actually cited. Door A = ordinary pages win (or the brand is already cited), so the brand's own page can compete; door B = at least two thirds of the cited sources are best-of lists, directories or marketplaces, so presence on those third-party pages is the lever, not rewriting the own page. Reported per question and per buyer-journey stage, with the share of door-B questions (pct_B). CitePulse never claims a brand is absent from a page; it shows which pages win.
- "Seen in k of n runs" (since 2026-09-05): a single run is a snapshot, not a state, because cited sources turn over by roughly two thirds day to day. Monitoring repeats the same frozen questions and reports each of the brand's own cited pages as seen in k of the last n comparable runs (same frozen question set, same measurement version MV-1), with the engines that cited it.
- Action Center and fix experiments (since 2026-09-05, monitoring plans): the client marks a fix as done (or describes any change; CitePulse also detects page changes by content hash). The change gets a date and a baseline of up to 3 earlier runs of the same series, none older than 60 days. Later runs are compared per question and engine with three states: rose beyond the engine's noise band, fell beyond it, or within normal variation. "Observed" requires at least 2 comparable runs, 6 repetitions per cell and 14 days after the change. All copy is correlational ("after this change"), never causal.
- Instrument-reliability panel NOISE-1.0 (since 2026-09-05): every day CitePulse asks the four engines the same five buying questions about citepulse.ai, three times each (60 calls), and publishes per engine the noise band (median day-over-day change in citation share per question), repetition agreement (share of question-days where three repetitions agreed) and day-to-day source overlap (median Jaccard of cited URLs), plus days measured. Below 14 measured days an engine uses a stated default band of 0.34. Public, machine-readable: https://api.citepulse.ai/api/public/noise (JSON, refreshed daily) and https://citepulse.ai/methodology#noise. Nothing from the panel enters any score.
- Questions and cadence (since 2026-09-06): every audit generates five buying questions grounded in the client's own site and freezes them as a comparable series; every monitoring plan measures weekly (Shield, Protect and Scale alike), asking each question three times per engine, and subscribers add their own buying questions on top (Shield up to 10, Protect up to 20, Scale up to 30 per client domain). That is roughly 580 measured answers per month on Shield (15 questions × 3 engines × 3 repetitions × weekly runs), about 1,080 on Protect and about 1,500 per client domain on Scale. A question added later is reported only against the runs that asked it, never as "missing" in earlier runs.
- AI bot logs (since 2026-09-06, monitoring plans): which pages the AI crawlers actually read (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, Googlebot, Bingbot, Applebot and others), from the client's own Cloudflare zone (read-only analytics token, encrypted at rest) or an uploaded log file; only day × crawler × path aggregates are kept, never IP addresses. Crossed with the pages the engines cite (read and cited / read, not cited / cited, rarely read), and with robots.txt. Correlation only.
- SEO × AI (monitoring dashboard): the client connects Google Search Console and GA4 (read-only, disconnectable) to see where Google ranks a page that AI does not cite and the reverse, and which sessions and conversions already arrive from AI assistants.
- Draft Studio, structural layer (since 2026-09-06): for the questions where a client's own page can win (door A), CitePulse assembles an FAQ block with FAQPage schema, a comparison-table skeleton (the client's column from approved facts, competitor cells left empty) and a visible last-updated line — deterministically, with no language model. Every sentence comes from the client's approved fact registry; missing facts are marked as gaps, never guessed. CitePulse does not write articles or marketing copy.
- Citation levels (since 2026-09-10): every answer is stored with three separate levels — the brand named in the answer text, the brand's domain among the cited sources, and an entity status (verified / unverified / namesake risk, because a same-name company on another domain can be the one mentioned). The headline still counts presence (named or cited); the levels are shown in the matrix and the report, and a namesake warning is explicit.
- Search trace (since 2026-09-14, monitoring plans): for every buying question CitePulse keeps what each engine searched for before answering (ChatGPT, Gemini and Claude report their search queries; Perplexity reports the pages it used), which pages came back and which were cited — read from the engines' own API responses, no browser automation. The dashboard shows per question whether the brand was cited, returned but not cited (Claude reports this; Perplexity and Gemini return only the pages they used), or not returned, plus the age of cited sources (Perplexity, Claude), where in the answer ChatGPT cites, and how many answer segments each source backs in Gemini. Forum threads (Reddit, Quora and similar) count as third-party pages.
- Data & export (since 2026-09-17, monitoring plans and the one-time Full Audit; not the free audit): every measurement downloads from the dashboard as CSV per dataset — runs, results per engine × question (presence levels), cited sources, engine searches, returned pages, approved facts, fact checks, AI crawler visits — or as one JSON bundle, with date, engine and question filters and a whole-range option, for the client's own analysis in a spreadsheet or their own AI assistant. Column names are plain English.
- Plain words in the dashboard (since 2026-09-17): the door A/B verdict reads "your own page can win" / "a third-party page wins", every before/after table carries a legend, and every action item says what to do.
- Microsoft Copilot is not measured: Microsoft offers no official API for the consumer Copilot (its documented Copilot APIs cover Microsoft 365 Copilot; Azure Grounding with Bing Search generates a model's own answer, not Copilot's). CitePulse does not automate browser sessions, shows Copilot as "not measured" in the question × engine matrix, and will add it when official programmatic access exists (checked 2026-09-06).
- Honesty: CitePulse sells measurement and fixes, not guaranteed citations. No engine publishes its source weights and citations shift day to day; llms.txt is treated as agent-readiness hygiene, not a citation factor. Methodology, versioned and public: CAIRS v2, Citation Share, DOORS-1.0, NOISE-1.0, measurement version MV-1 — https://citepulse.ai/methodology
- Pricing: free mini-audit (no signup or card); Full Audit $39 one-time + $9 per extra service area; monitoring Shield $89/mo, Protect $169/mo, Scale $349/mo (white-label for agencies, up to 10 client domains, API). Enterprise is custom. Machine-readable pricing for agents: https://citepulse.ai/pricing.md
- Markets: English-canonical (global) with a Polish edition at /pl/. Customers: SMBs, in-house marketing teams and agencies.

## Pages

- [Home](https://citepulse.ai/): offering, four-layer score, Fact Integrity, sample report, pricing, FAQ
- [AI Visibility Monitoring (the product)](https://citepulse.ai/monitoring): continuous monitoring for B2B brands: baseline on day one, the same buyer questions every week on ChatGPT, Perplexity and Gemini (Grok on Protect and Scale; Claude only in the one-time Full Audit), presence of the brand in the answers and its changes, competitor visibility, cited sources, alerts when a competitor takes the answer, before/after against the engines' noise band; plans $89 / $169 / $349 a month
- [Pricing](https://citepulse.ai/pricing): plans ($89 / $169 / $349 a month, weekly runs; free audit; $39 one-time Full Audit), limits per plan, extras ($9 per extra area, $30 per client domain above 10 on Scale), billing (USD only, VAT invoice from GeoFlow Solutions), cancellation. Polish: https://citepulse.ai/pl/pricing
- [For agencies](https://citepulse.ai/agencies): white-label client reports, 10 client domains on Scale, per-client badges, API endpoints, weekly workflow, what is not included. Polish: https://citepulse.ai/pl/agencies
- [Fact Integrity](https://citepulse.ai/fact-integrity): how wrong prices and facts in AI answers are detected (approved registry, ai_answer / web_crawl / partner_listing checks, match / partial_match / conflict / missing), a documented example from our own domain, limits
- [Free mini-audit](https://citepulse.ai/audit): audit form — website + service areas, an AI Reputation Score in ~30 seconds, PDF report by email
- [Full Audit configurator](https://citepulse.ai/configure): the paid $39 one-time Full Audit — adds Claude, per-area competitor breakdown and a monitoring credit
- [Free tools](https://citepulse.ai/tools/): AI-bot access checker, AI-readability check, Organization schema generator, llms.txt generator
- [Developers: API & MCP](https://citepulse.ai/developers): self-serve REST API and read-only MCP servers to run AI-visibility audits, with authentication, endpoints, examples, limits and pricing (no enterprise contract)
- [AI Visibility Setup (done-for-you)](https://citepulse.ai/services/ai-visibility-setup): Starter / Growth / Pro one-time fix engagements
- [Blog](https://citepulse.ai/blog/): research and practical guidance on generative engine optimization (GEO), answer engine optimization (AEO) and AI citability
  - [You Can't Buy AI Citations (we tested it on ourselves)](https://citepulse.ai/blog/you-cant-buy-ai-citations): 351 AI answers across five engines, zero cited citepulse.ai; why citations are earned and measured, not bought, with the Martinez 2026 survey of 45 GEO studies
  - [How much do AI answers change from day to day? Measuring the noise before you trust a before/after](https://citepulse.ai/blog/how-much-do-ai-answers-change-day-to-day): cited sources turn over by roughly two thirds day to day; why a before/after needs a noise band; how the public NOISE-1.0 panel works; what "seen in k of n runs" means
  - [Why AI cites small brands over big ones](https://citepulse.ai/blog/why-ai-cites-small-brands-over-big-ones): a small brand beat three fashion giants 29 mentions to 2 in AI search; Anthropic's new browser tool reads a page's structure, not its pixels; a CitePulse benchmark of 109 companies shows a median AI visibility of 13/100 with more than a third scoring zero
  - [Citation share: the new share of voice for SEO, GEO and AEO](https://citepulse.ai/blog/citation-share-the-new-share-of-voice): why classic share of voice is eroding and how citation share replaces it, with a 25-company benchmark (median AI visibility 12/100)
  - [Why ChatGPT doesn't recommend your business](https://citepulse.ai/blog/why-chatgpt-doesnt-recommend-your-business): the six technical blockers and how to fix them
  - [llms.txt: what it is, who actually reads it, and whether you need one](https://citepulse.ai/blog/llms-txt-what-it-is-and-whether-you-need-one): why 97% of files get zero requests and no study shows a citation effect
  - [What is Answer Engine Optimization (AEO)](https://citepulse.ai/blog/what-is-answer-engine-optimization): what AEO is and why businesses are invisible to AI
  - [What is Generative Engine Optimization (GEO) and how to track it](https://citepulse.ai/blog/what-is-generative-engine-optimization-and-how-to-track-it)
  - [The technical checklist for AI citability](https://citepulse.ai/blog/technical-checklist-for-ai-citability): robots.txt, JavaScript, schema.org, sitemap and llms.txt
  - [How AI assistants decide which businesses to recommend](https://citepulse.ai/blog/how-ai-assistants-decide-what-to-recommend): citation frequency, source presence and competitor displacement
  - [How to check if AI cites your brand](https://citepulse.ai/blog/how-to-check-if-ai-cites-your-brand)
  - [How to measure AI visibility across ChatGPT, Perplexity, Gemini and Claude](https://citepulse.ai/blog/how-to-measure-ai-visibility-across-chatgpt-perplexity-gemini-claude)
  - [How to check and improve your visibility in Perplexity](https://citepulse.ai/blog/how-to-check-and-improve-your-visibility-in-perplexity): checking, metrics, improving and monitoring Perplexity AI visibility
  - [Passage extractability: why AI quotes some pages and skips others](https://citepulse.ai/blog/passage-extractability-why-ai-quotes-some-pages): how AI lifts self-contained passages, what Google's patents suggest, and how to make pages extractable
  - [How to write content Google's AI Overviews can extract](https://citepulse.ai/blog/how-to-write-content-ai-overviews-can-extract): what "built to be extracted" means, four moves to make sections quotable, and how to measure whether AI cites you
  - [Best AI visibility monitoring tools compared](https://citepulse.ai/blog/best-ai-visibility-monitoring-tools-compared): CitePulse, Otterly, Profound, Peec and Knowatoa
- [Alternatives (honest, sourced comparisons)](https://citepulse.ai/alternatives/): CitePulse compared head-to-head with the main AI-visibility tools, including where each rival is the better pick
  - [CitePulse vs Ahrefs Brand Radar](https://citepulse.ai/alternatives/ahrefs-brand-radar)
  - [CitePulse vs Evertune](https://citepulse.ai/alternatives/evertune)
  - [CitePulse vs Frase](https://citepulse.ai/alternatives/frase)
  - [CitePulse vs LLM Pulse](https://citepulse.ai/alternatives/llm-pulse)
  - [CitePulse vs Otterly.AI](https://citepulse.ai/alternatives/otterly)
  - [CitePulse vs Profound](https://citepulse.ai/alternatives/profound)
  - [CitePulse vs Peec AI](https://citepulse.ai/alternatives/peec-ai)
  - [CitePulse vs Knowatoa](https://citepulse.ai/alternatives/knowatoa)
- [Research: citepulse.ai measured on itself](https://citepulse.ai/research/citepulse-ai): dated study, 6 buying questions × 5 engines × 3 repetitions/day since 2026-09-05, every cited URL, CSV/JSON downloads; result so far: 0 mentions, 0 domain citations; the pages the engines cite instead
- [Public real report (no form)](https://citepulse.ai/public/report/citepulse.ai): the Full Audit of citepulse.ai exactly as delivered on 2026-08-11; not demo data
- [Methodology](https://citepulse.ai/methodology): how every number is computed — detection rules, Citation Share denominator, doors A/B (DOORS-1.0), the live noise panel (NOISE-1.0), honest limits
- [Public noise data (JSON)](https://api.citepulse.ai/api/public/noise): NOISE-1.0 per engine, refreshed daily
- [Security](https://citepulse.ai/security): infrastructure, encryption, sub-processors, responsible disclosure
- [Polish edition](https://citepulse.ai/pl/): Polish-language version of the site

## Legal

- [Terms of Service](https://citepulse.ai/legal/terms): customer agreement; no guarantee of results; as-is/as-available; limitation of liability capped at fees paid in the prior 12 months; auto-renewal; B2B vs consumer; API terms
- [Privacy Policy](https://citepulse.ai/legal/privacy): GDPR + US; data collected, lawful basis, retention, data-subject rights, international transfers (SCCs)
- [Data Processing Addendum (DPA)](https://citepulse.ai/legal/dpa): GDPR Art. 28 processor terms, sub-processor list, SCC reference
- [Cookie Policy](https://citepulse.ai/legal/cookie): strictly necessary storage, currency preference, privacy-first consent approach
- [Refund & Cancellation Policy](https://citepulse.ai/legal/refund): how to cancel subscriptions, one-time audit refund stance, Stripe billing, EU withdrawal right
- [Product Disclaimer](https://citepulse.ai/legal/disclaimer): results informational, no citation guarantee, AI variability, not professional advice
- Status: the legal pages above are published and in force (operator: GeoFlow Solutions Marcin Gamża, Polish law; sales live since 2026-07-14). A review by licensed counsel is planned; this line will say so when it is done.

## Company

- Owner: GeoFlow Solutions Marcin Gamża — "Your Application Intelligence Partner". Brand form: "CitePulse.ai — by GeoFlow Solutions".
- Disambiguation: citepulse.ai is NOT connected to citepulse.io (a different company's product with a similar name). Facts, prices and figures on this domain describe citepulse.ai only.
- Contact: contact@geoflow.solutions
- LinkedIn: https://www.linkedin.com/company/citepulse-ai
- Product update, September 2026: https://citepulse.ai/blog/what-changed-in-citepulse-september-2026