What DataFawn stores, and why.

Data we store

• Account data: your email address, optional name, timezone, and workspace memberships.
• Search Console data: the performance rows and aggregates imported for the properties you select.
• Website evidence: pages fetched during a bounded crawl, including titles, headings, canonical signals, and internal links.
• Product data: keywords, competitors, SERP snapshots, opportunities, recommendations, experiments, and results.
• Operational data: an append-only activity log, provider usage and cost counters, and job status.

Google data use

DataFawn requests the read-only Search Console scope and uses that data only to produce analysis and recommendations inside your workspace. It is never used to train models, is never sold, and is never shared with other customers. You can disconnect the integration at any time, and deleting your workspace revokes the credential at Google.

Model providers and training

DataFawn uses the OpenAI API and the gpt-5.6-luna model for every generative, structured-output feature. This is the only production LLM route. Extraction, classification, second-view checks, synthesis, explanations, recommendations, and drafting all use the same OpenAI model and the same OpenAI usage budget.

OpenRouter is not integrated into the current product and receives no DataFawn data. No DeepSeek, Mimo, or other third-party LLM is called by the application. The separate OpenAI embedding model described below is a non-generative similarity service, not an alternative LLM route.

Other model and API services

DataFawn also uses text-embedding-3-small through OpenAI's embeddings API for one narrow purpose: deterministic entity near-duplicate matching. For that operation, DataFawn sends batches of canonical entity names, entity types, and stored aliases. It does not send full page content, Search Console rows, prompts, recommendations, or workspace questions to the embeddings endpoint. The application compares the returned vectors with a fixed similarity threshold; an LLM does not make the merge decision.

Jina Reader and Autom.dev are retrieval APIs, not configured DataFawn LLM routes. Jina Reader receives selected URLs and returns readable page content. Autom.dev receives tracked keyword and search parameters and returns live SERP snapshots. Google Search Console receives the authorization scope needed for its read-only API access. Metrics, thresholds, quotas, deduplication, scoring, and merge thresholds are calculated deterministically by DataFawn.

DataFawn does not train models on your data and does not grant a provider permission to train on your behalf. OpenAI API processing, security, retention, and any applicable data-control terms are also governed by OpenAI's then-current API terms and privacy documentation. DataFawn uses the API service, not the consumer ChatGPT product.

What we send to OpenAI

We send the minimum workspace evidence needed for the specific action you request or the scheduled feature that is enabled. A request contains the Luna model name, an instruction describing the task, a JSON-output requirement, and a structured evidence payload. The payload can contain the following categories:

• Content extraction: normalized text from the selected page, truncated to the configured 100,000-character analysis limit. Luna extracts headings, topics, entities, format, intent signals, CTA patterns, and supported entity relations.
• Content refresh: the page URL, refresh objective, optional brand guidelines, normalized current page content (up to the same analysis limit), and the supplied evidence snapshot. Evidence can include source IDs, source types, capture dates, summaries, and structured SERP, crawl, Search Console, or opportunity data.
• Recommendations and plans: the selected opportunity, page, competitor, crawl, SERP, Search Console, alert, experiment, and measurement evidence needed to produce a recommendation or weekly action plan. The model receives the evidence references and metrics, not an unrestricted database export.
• Ask DataFawn and explanations: the question or requested explanation, the relevant date range, and the workspace evidence selected by the deterministic query. Drop analysis also includes the supplied decline evidence and fixed metrics.
• Growth Digest: bounded competitor and first-party Search Console signals such as signal type, keyword, entity, previous and current positions, clicks, impressions, dates, summaries, and calculated metrics. The digest uses two independent Luna classification passes and one Luna synthesis pass; all three go to OpenAI.
• Entity coverage and experiment explanations:deterministic entity-gap candidates, evidence references, fixed experiment metrics, classifications, confidence values, and outcome limitations needed for the explanation.

URLs, page text, query terms, competitor names, entity names, business objectives, and brand guidelines may contain personal or confidential information if you put that information on your site or in your workspace. Do not submit secrets or information you are not authorized to process. DataFawn does not intentionally send passwords, session cookies, OAuth access or refresh tokens, payment card details, provider API keys, or unrelated workspaces to OpenAI.

What we send to other processors

• Google Search Console: the read-only authorization and selected property scope needed to import search-performance data.
• Autom.dev: tracked keywords and search settings needed to retrieve live SERP snapshots.
• Jina Reader: URLs selected for analysis, so readable page content can be fetched. The resulting content can later become evidence sent to OpenAI when an AI feature uses it.
• Resend: recipient email addresses and the content required for sign-in links, alerts, and product email.
• Paddle: checkout, subscription, and billing details required to process payment as merchant of record. DataFawn does not receive or store full payment-card numbers.
• Linux VPS and object storage: application data, database records, queued jobs, crawl artifacts, and encrypted credentials required to operate DataFawn.

What DataFawn stores after an AI request

DataFawn stores the validated result in your workspace when the request succeeds. It also records the AI purpose, provider (OpenAI), model (gpt-5.6-luna), prompt version, input hash, input and output token counts, estimated cost, status, and failure code where applicable. Evidence snapshots are preserved so a recommendation can be audited against the information available when it was generated. The raw prompt is not displayed to other customers.

Raw crawl HTML follows the configured raw-evidence retention window (30 days by default), raw provider responses follow the configured provider-response window (90 days by default), and public reports follow the configured public-report window (180 days by default). Workspace artifacts and audit records remain while the workspace exists unless a shorter retention rule applies. These windows can be changed by the deployment operator.

Third-party processors

• Google Search Console API — reads your search performance data with the read-only scope you grant.
• Autom.dev — retrieves live search-result snapshots for keywords you track.
• Jina Reader — fetches readable page content for pages under analysis.
• OpenAI Embeddings API — text-embedding-3-small — creates embeddings from normalized entity names for deterministic near-duplicate matching.
• OpenAI API — gpt-5.6-luna — the sole generative AI provider for content extraction, classification, synthesis, explanations, and recommendations.
• Resend — sends sign-in links and product email.
• Paddle — processes payments as merchant of record and holds your billing details.
• Linux VPS hosting — runs the application, PostgreSQL, Redis, and object storage.

Retention and deletion

Workspace data is kept while the workspace exists. Raw crawl HTML and raw provider responses are kept only as long as the configured retention window for that workspace. The default windows are:

These windows can be changed by the deployment operator, and shorter windows always win. A daily retention job applies the lifetimes idempotently, so re-delivery of a job never double-deletes.

Deletion workflow

A workspace owner can delete a workspace from the data and deletion page in workspace settings. Deletion cancels the subscription, revokes the Google connection, removes workspace data, and cannot be undone. If a provider revocation fails, the remaining identifiers are removed from DataFawn and you can finish removal from the provider side. Only an append-only audit entry and a deletion receipt of identifiers are retained.

Your rights

You can request access to, correction of, or export of the data DataFawn holds about you, and you can ask for deletion at any time. Where we process data because a workspace owner configured it, the owner can delete or export it through the product; we will also honor a direct request. Email support@datafawn.com and we will respond without undue delay.

Public sharing consent

Nothing is public by default. Public Win Card, Weekly Wrapped, and outcome links exist only after you create them — creating a link is your explicit consent to share that summary. Links use unguessable tokens, can expire, and can be revoked at any time; revoking a link removes the public view immediately. A public link never exposes payment details, credentials, or other workspaces.

Contact

For access, correction, export, or deletion requests, email support@datafawn.com.