# Selfwin Docs: complete documentation > Documentation for tracking AI visibility, interpreting Selfwin analytics, and turning lost prompts into approved actions. --- # Welcome to Selfwin > Understand what Selfwin tracks, what it calculates, and how its approval-gated AI employee turns evidence into work. - Product: Selfwin - Section: Get started - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/introduction - Keywords: AI visibility, generative engine optimization, GEO, Selfwin Selfwin tracks how brands appear in answers from AI systems. It collects answers for questions your customers are likely to ask, measures which brands and sources appear, and keeps the original evidence behind every number. Selfwin also has an AI employee. It reviews the tracking evidence, identifies lost prompts, drafts work that could improve them, and puts that work in an approval queue. It does not publish external work automatically. ## The product in three layers ### Track Selfwin runs active prompts against the model channels enabled for a project. A completed run can store: - The prompt, country, model channel, and underlying model identifier. - The complete answer text. - Brand mentions, occurrence counts, position, sentiment, and supporting snippets. - Retrieved URLs, domains, page titles, citations, and source classifications. - Search fanouts observed while the answer was produced. The answer is the base unit of the system. Dashboard metrics are summaries of those stored answers, not separate estimates. ### Understand Selfwin turns the collected evidence into four brand metrics: visibility, share of voice, sentiment, and position. It also measures how often sources are retrieved and cited, compares brands across models and topics, and runs separate perception studies against editable market attributes. Read [Metrics overview](./metrics) for the exact formulas. ### Act The Visibility Agent looks for prompts where your brand is absent, competitors lead, another site gets the credit, or your own source is used without your brand being named. It creates evidence-backed actions and drafts for review. Nothing leaves Selfwin until a person approves it. Approval changes the workflow state; it does not silently publish to a website, community, or third party. ## What Selfwin can collect Provider-backed collection supports OpenAI, Anthropic, Gemini, and Perplexity when the corresponding project credentials are configured. Browser collection is a separate option for supported consumer-facing model experiences and can capture observable search activity that an API response may not expose. Availability depends on the credentials and browser infrastructure configured for the project. The Settings page is the source of truth for what can run. ## Start here 1. Follow the [Quickstart](./quickstart) to create a project and run the first prompts. 2. Learn [how tracking works](./tracking) before comparing numbers across tools. 3. Use the [Overview](./overview) to understand the result. 4. Open the [Visibility Agent](./visibility-agent) when you are ready to turn a gap into an action. ## Documentation and machine access Every documentation page has an equivalent Markdown URL ending in `.md`. The documentation index is available at [`/llms.txt`](./llms.txt), and the complete corpus is available at [`/llms-full.txt`](./llms-full.txt). These docs describe implemented behavior. They intentionally exclude planned or speculative features. --- # Quickstart > Create a project, choose prompts and model channels, run tracking, and review your first visibility result. - Product: Selfwin - Section: Get started - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/quickstart - Keywords: Selfwin setup, first tracking run, onboarding You can reach a useful first result with one project, a focused prompt set, and at least one configured model channel. ## 1. Create a project Open **Start tracking** and enter the public domain for the brand you want to measure. Selfwin reads the site’s public metadata and uses it to prepare a project brief. Review the proposed: - Project and brand name. - Primary domain. - Brand description. - Products, audience, category, and positioning. The brief gives prompt generation and the Visibility Agent shared context. Correct it before generating a large prompt set. ## 2. Confirm the brand identity In Settings, check the tracked brand name, aliases, and domains. Use aliases for normal spelling variants. Use a custom matching expression only when literal names are not precise enough. Accurate identity matters because Selfwin reuses the same matching rules for visibility, share of voice, position, sentiment, and historical recalculation. ## 3. Add prompts Open **Prompts** and either generate suggestions, add questions manually, or import a CSV. A prompt should resemble a real question, for example: > What is the best no-code app builder for an internal operations tool? Keep prompts at 200 characters or fewer. Assign a two-letter country code, a topic, and tags where useful. Suggested prompts do not run until you accept them. ## 4. Configure collection Open **Settings → Model tracking** and enable the channels for which credentials or browser collection are available. Then choose: - Daily or weekly tracking. - Browser or provider API collection for scheduled runs. - The project time zone. Browser collection and API collection can produce different answers. Treat them as separate model channels when comparing trends. ## 5. Run tracking Select **Run tracking** from the application header. Selfwin creates one answer per active prompt and selected engine for the project’s current run date. Existing successful prompt-engine-day combinations are skipped, which prevents accidental duplicate data. The run history reports completed, skipped, and failed items. A failed answer is not counted as an analyzed response in visibility metrics. ## 6. Read the first result Open **Overview** after the run finishes. Check: 1. The number of answers collected. 2. How many active prompts are represented. 3. Your visibility and the leading competitor. 4. Which sources appear most often. 5. The recent answers behind those numbers. A first run is a snapshot, not a trend. Continue collecting on a consistent prompt and model set before treating small changes as a durable movement. ## 7. Run the Visibility Agent When tracking data exists, open **Visibility Agent** and start an analysis. The agent classifies the clearest gaps, records its evidence, and creates draft actions. Review those drafts in the [Action queue](./actions). Approving an action does not publish it externally. ## Next steps - Learn the [metric formulas](./metrics). - Organize a larger prompt set with [topics and tags](./prompts). - Verify the evidence in [Chats](./chats). - Connect [Slack or an MCP client](./integrations). --- # How tracking works > Learn how Selfwin runs prompts, separates model channels from model versions, stores evidence, and handles retries and duplicate runs. - Product: Selfwin - Section: Get started - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/tracking - Keywords: AI answer collection, model channels, browser tracking, API tracking Tracking is the process of sending each active prompt to each enabled model channel and saving the resulting answer and evidence. ## The collection matrix The planned work for a run is: ```text active prompts × enabled model channels ``` Each successful answer is unique for one prompt, one engine, and one project-local date. Running tracking again on the same day skips successful combinations that already exist. Failed combinations can be retried. This makes daily collection idempotent while keeping failures visible. ## Model channel and model version Selfwin stores two model identities when available: - **Model channel** is the stable surface, such as ChatGPT UI, OpenAI API, Gemini UI, or Gemini API. - **Model ID** is the particular underlying model version reported for that answer. Model channels keep a trend continuous when a provider changes the model behind a consumer product. The model ID preserves the technical detail needed to explain a change. ## Browser and API collection Selfwin supports provider APIs and separate browser-based collection. ### Provider API API collection uses the project’s configured provider credentials. It is generally direct and structured, but the answer or sources can differ from what a logged-out consumer sees in a web product. ### Browser collection Browser collection uses a fresh session for supported model websites. It can preserve visible sources and observed background searches. Production scheduling requires configured cloud-browser infrastructure; local Chrome is a development fallback. Do not merge browser and API channels when you need a like-for-like comparison. ## What a completed answer contains Depending on the model response, Selfwin stores: - Full answer text and raw provider metadata. - Mentioned tracked brands and newly observed brands. - The number of textual occurrences for each brand. - First-mention position, sentiment, and a supporting snippet. - Retrieved URLs and domains. - Citation counts and citation positions. - Observed fanout queries. - Response features, maps, ads, and products when present in the collector output. ## Analysis happens after collection After an answer is stored, Selfwin analyzes it against the project’s brand definitions. Literal tracked names and aliases are matched case-insensitively. Custom matching expressions are case-sensitive. The result is recalculable. Changing a brand name, alias, domain, or matching expression can trigger historical analysis so old answers follow the current brand identity. ## Scheduling Projects can use daily or weekly tracking. Scheduled runs use the project’s time zone and selected collection method. Manual tracking is available from the application header. The scheduled Visibility Agent and Slack brief are separate jobs from answer collection. They read stored tracking evidence; they do not replace it. ## Coverage and comparability Two visibility percentages are comparable only when their underlying populations are comparable. Check: - Date range. - Active prompts represented. - Model channels. - Countries. - Topics and tags. - Failed or newly added prompt-engine combinations. The Overview shows response and prompt coverage so a thin data set is not mistaken for a complete market measurement. --- # Metrics overview > Exact definitions for visibility, share of voice, sentiment, position, retrieval, and citation metrics in Selfwin. - Product: Selfwin - Section: Understand results - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/metrics - Keywords: visibility formula, share of voice, sentiment, position, citation rate Selfwin calculates metrics from completed answers inside the selected filters. Failed and pending answers are not part of the denominator. ## Brand metrics ### Visibility Visibility is the percentage of analyzed answers in which a brand appears at least once. ```text visibility = answers mentioning the brand ÷ all analyzed answers × 100 ``` If a brand appears in 31 of 100 answers, its visibility is **31%**. Repeating the brand five times in one answer still contributes one visible answer. ### Share of voice Share of voice measures the brand’s portion of all occurrences of tracked brands. ```text share of voice = occurrences of this brand ÷ occurrences of all tracked brands × 100 ``` If the selected answers contain 20 tracked-brand occurrences and 5 belong to your brand, your share of voice is **25%**. Visibility and share of voice answer different questions. Visibility measures reach across answers. Share of voice measures how much of the brand conversation you occupy. ### Sentiment Sentiment is the average 0–100 contextual score for answers in which the brand is detected. Higher is more positive. Answers without a usable sentiment score are excluded from the sentiment denominator. ### Position Position is the average order of the brand’s first appearance among detected brands in an answer. Lower is better. - Position 1 means the brand appears first. - Position 2 means one detected brand appears before it. - Answers where the brand is absent do not receive a position. ### Change indicators Changes compare the selected period with the immediately preceding period of the same length. For position, a lower number is an improvement. For visibility, share of voice, and sentiment, a higher number is an improvement. ## Prompt metrics Prompt rows apply the same brand formulas to answers generated by one prompt. Prompt volume is a 1–5 relative bucket when data is available. It is an organizational demand signal, not a count of Selfwin answers. ## Source metrics Source metrics describe retrieved and explicitly cited URLs separately. ### Retrieved The percentage of analyzed answers that retrieved at least one URL from the source. ```text retrieved = answers retrieving the source ÷ all analyzed answers × 100 ``` ### Retrieval rate The average number of retrieved URL occurrences from the source per analyzed answer. ```text retrieval rate = source retrievals ÷ all analyzed answers ``` This value can be greater than 1. ### Citations The total number of explicit citation occurrences assigned to the source. ### Citation share The source’s portion of all explicit citations in the selected answers. ```text citation share = source citations ÷ all citations × 100 ``` ### Citation rate The average number of explicit citations when the source was retrieved. ```text citation rate = source citations ÷ answers retrieving the source ``` This value can also be greater than 1. ## Brand visibility and source visibility A brand can be named without its website being used. Its website can also be retrieved or cited without the brand being named. Selfwin keeps these facts separate because the gap between them is actionable. For example, frequent citations with weak brand visibility can indicate that models trust the content but do not connect it strongly enough to the brand name. --- # Overview dashboard > Read visibility trends, brand rankings, source distribution, collection coverage, and recent answers from one page. - Product: Selfwin - Section: Understand results - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/overview - Keywords: Selfwin dashboard, visibility trend, brand ranking Overview is the main analytical summary for a project. Every card uses the same selected answer population unless the card explicitly says otherwise. ## Filters The shared filter bar can narrow results by: - Preset or custom date range. - Competitor. - Model channel. - Topic. - Tags, with AND or OR matching. - Country. Options appear when relevant project data exists. Use identical filters when comparing numbers between pages. ## Visibility trend The trend chart plots your brand and the leading competitors over time. Switch between daily, weekly, and monthly resolution without changing the underlying date range. Empty calendar dates remain visible in a daily series. A zero means analyzed answers exist but the brand was absent. No point means there was no analyzed answer for that date and scope. The footer shows the number of responses and how many active prompts are represented. Use this coverage before drawing conclusions from a movement. ## Brand ranking The table ranks your brand and the leading competitors by visibility. It also shows: - Share of voice. - Sentiment. - Position. - Change from the preceding equal-length period. Your own brand remains identifiable even when it is not the current leader. ## Source distribution The domain list shows the most frequently retrieved sources in the selected answers. The adjacent distribution groups those retrievals by source type, such as Editorial, Corporate, Institutional, UGC, Reference, Competitor, Related, Own website, or Unclassified. Open the [Sources](./sources) area for complete ratios, URLs, trends, classifications, and bookmarks. ## Recent chats Recent chats are the actual collected answers behind the dashboard. Use **Brand mentions only** to hide answers where no tracked brand was detected. An answer with no brand mention is still valid tracking data. It remains part of the visibility denominator and helps explain a zero-visibility prompt. ## Empty states Overview shows **No tracking data yet** only when there are no completed analyzed answers in scope. It does not use your own brand’s visibility as a proxy for whether tracking ran. If the dashboard is unexpectedly empty: 1. Reset filters. 2. Check Run history for completed answers. 3. Confirm active prompts and configured channels in Settings. 4. Check whether the selected date range includes the run date in the project time zone. --- # Chats and answer evidence > Inspect the collected AI answers that form the denominator and evidence for every Selfwin analytics result. - Product: Selfwin - Section: Understand results - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/chats - Keywords: AI answers, chat evidence, citations, model responses A chat is one stored answer for a prompt, engine, and run date. Chats are the evidence layer beneath every dashboard metric. ## Chat list The chat list can be filtered by the shared analytical dimensions and searched through the application’s global search. Each row identifies the prompt, model channel, run date, detected brands, and relevant performance details. Answers with no detected brand remain visible. They are important because they contribute to the visibility denominator. ## Chat detail Open a chat to inspect: - The complete prompt and answer. - Model channel and underlying model ID when available. - Country and collection date. - Detected brands with occurrence count, first position, sentiment, and snippet. - Retrieved and cited sources. - Observed fanout queries. - Structured response features captured by the collector. Use the original answer when a metric looks surprising. The stored evidence should make it possible to explain why a brand was counted, where it appeared, and which sources influenced the answer. ## Mention analysis One mention record is stored per detected brand per answer, but it includes the number of textual occurrences. This supports two different calculations: - Visibility counts the answer once. - Share of voice counts all occurrences. Position records the order of the brand’s first appearance among detected brands. Sentiment records contextual tone on a 0–100 scale when analysis produces a score. ## Sources and citations A retrieved source is not necessarily an explicit citation. Selfwin stores both the retrieved URL and the number of times it was explicitly cited where the collector can distinguish them. Source pages aggregate this evidence by domain or URL. See [Sources, domains, and URLs](./sources). ## Failed answers Pending and failed collection attempts can retain status and an error message, but they do not enter completed-answer metric denominators. Run history shows whether a thin data set came from skips, failures, or a small prompt set. ## Externally collected answers An authenticated MCP client can submit an externally collected anonymous answer for analysis. Selfwin processes it through the same brand, citation, fanout, and metric pipeline as native tracking evidence. --- # Fanout queries > See the searches and retrieval targets an AI system used while composing a tracked answer. - Product: Selfwin - Section: Understand results - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/fanouts - Keywords: query fanouts, AI search queries, retrieval evidence Some AI systems perform background searches before composing an answer. Selfwin calls those searches **fanout queries**. Fanouts help explain why a particular page or competitor entered an answer. They can also reveal the language an AI system uses to decompose a customer’s question. ## Observed evidence only Selfwin stores fanouts that were visible in the browser interface, network activity, or provider grounding metadata. It does not ask another language model to guess which searches probably happened. Each stored fanout can include: - Query or retrieval target. - Type, such as search, shopping, or synthetic where the collector identifies it. - Order within the answer. - Collection source. - Evidence describing where it was observed. ## Distinct queries and occurrences The same normalized query is stored once per chat. Across many chats, Selfwin can distinguish: - **Distinct queries:** Different query texts observed in the selected scope. - **Occurrences:** The total number of chats in which those queries appeared. A repeated query is valuable evidence that a topic or source pattern is stable rather than incidental. ## Grouping The Fanouts page can organize evidence by the prompt that triggered it. Prompt detail pages show the corresponding granular evidence for one question. Use filters to keep the model, country, topic, tag, and date population aligned with the dashboard result you are investigating. ## How the Visibility Agent uses fanouts The agent clusters observed searches and associates them with winning URLs when evidence exists. It checks whether your brand is present, who owns the source, and whether the result appears winnable. Fanouts can support actions such as: - Retargeting an existing page to answer a recurring query. - Creating a comparison or category page. - Addressing a community thread that models repeatedly retrieve. - Fixing indexing when an appropriate owned page exists but is not retrieved. Fanouts are evidence for an action, not proof that executing the action will change a future model answer. --- # Sources, domains, and URLs > Analyze what AI models retrieve and cite, classify sources, inspect trends, and find gaps between source and brand visibility. - Product: Selfwin - Section: Understand results - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/sources - Keywords: AI sources, domains, URLs, retrieval rate, citation rate Sources are the pages AI systems retrieve or cite while producing tracked answers. Selfwin separates domain-level patterns from individual URL performance. ## Domains and URLs - **Domains** combine every captured URL from one hostname. - **URLs** preserve performance for one exact page. Both views use the same selected chat population. See [Metrics overview](./metrics) for the exact source formulas. ## Source table The table reports: - Retrieved percentage. - Retrieval count and retrieval rate. - Citation count, citation share, and citation rate. - Source classification. - First and last observed dates where available. - Whether the source is bookmarked. Use **Top**, **New**, **Trending**, and **Losing** to change how rows are ordered or filtered. Movers compare equivalent periods rather than treating raw lifetime totals as a trend. ## Source classifications Built-in source types include: - Editorial. - Corporate. - Institutional. - User-generated content. - Reference. - Competitor. - Related. - Other. - Own website. - Unclassified. You can create project-owned classifications and override the classification of a domain or URL. Overrides affect organization and analysis; they do not change the original captured URL. ## Bookmarks Bookmark a domain or URL to keep it in a focused source list. Bookmarks are scoped to the project and to either domain or URL view. ## Detail pages A detail page shows the source’s core ratios, retrieval trend, related prompts, chats, and brands. URL details can include the stored page title and captured content when available. This supports questions such as: - Which prompts repeatedly retrieve this site? - Did retrieval rise before or after a visibility change? - Is a competitor named when this source appears? - Does an owned page get used without the brand being named? ## Gap analysis Selfwin compares your brand presence with competitor presence inside answers using a source. The resulting gap score is a Selfwin prioritization signal; it is not a probability or a guaranteed ranking improvement. Source visibility and brand visibility remain separate. Being cited does not automatically mean the model names the brand. --- # Prompts, topics, and tags > Build the questions Selfwin tracks, organize them, import them in bulk, and understand prompt-level performance. - Product: Selfwin - Section: Configure tracking - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/prompts - Keywords: AI prompts, prompt tracking, topics, tags, CSV import Prompts are the questions Selfwin sends to AI model channels. A useful prompt represents a question a potential customer, evaluator, or researcher might genuinely ask. ## Prompt states - **Active** prompts are included in future tracking runs. - **Suggested** prompts are saved ideas waiting for acceptance or rejection. - **Archived** prompts stop running but keep their historical answers. Deleting a prompt is different from archiving it. Deletion also removes its associated answers, mentions, citations, and fanouts. ## Add prompts You can add one or many prompts manually, generate suggestions from the project brief, or import CSV data. A prompt must contain 1–200 characters. A single request can add up to 500 prompts; exact duplicates already in the project are skipped. CSV import recognizes a prompt column and can attach supported organizational fields. Review imported rows before starting a broad tracking run. ## Countries Each prompt stores a two-letter country code. The default is `US`. Country becomes an analytical dimension in Overview and Insights, which lets the same topic be compared across markets. Country identifies the configured market for the prompt. It does not guarantee that every provider response is physically generated from an IP address in that country. ## Topics A prompt belongs to zero or one topic. Topics are useful for broad areas such as Pricing, Integrations, Reputation, or Use cases. Deleting a topic does not delete its prompts. Those prompts move to **No topic**. ## Tags and tag groups Tags cut across topics. A prompt can have multiple tags, and user-managed tags can be grouped and colored. Selfwin also maintains system dimensions for: - **Branding:** Branded or Non-branded. - **Intent:** Informational, Commercial, Transactional, or Navigational where detected. System tags are generated from the prompt language and protected from ordinary editing. Shared filters support tag AND/OR logic. ## Prompt performance The Prompts table shows the selected brand’s: - Visibility. - Sentiment. - Position. - Mentioned brands. - Relative volume bucket. - Topic, country, tags, and added date. Open an active prompt for a focused page containing its metrics, recent answers, and observed fanout evidence. ## Bulk management Select multiple prompts to: - Activate or archive them. - Assign a topic. - Assign tags. - Delete them. Use archive when you may need the historical series later. Use delete only when the prompt and all of its evidence should be removed. --- # Competitors > Define tracked brands precisely, review suggestions, and recalculate historical answers when identity rules change. - Product: Selfwin - Section: Configure tracking - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/competitors - Keywords: competitor tracking, brand aliases, brand matching Competitors are brands measured beside your own brand. They share the same visibility, share-of-voice, sentiment, and position calculations. ## Tracked and display identity Each brand can contain: - **Display name:** The label shown in tables and charts. - **Tracked name:** The primary literal name searched for in answers. - **Aliases:** Alternative spellings or names. - **Domains:** One or more websites associated with the brand. - **Matching expression:** An optional case-sensitive expression for precise cases. - **Color:** The brand’s chart color. Literal tracked names and aliases match without case sensitivity. Custom expressions remain case-sensitive so their behavior is explicit. ## Suggestions Selfwin can propose brands found repeatedly in collected answers. A candidate must appear across distinct answers rather than being repeated several times in one answer. Suggestions remain separate from tracked competitors until accepted. You can accept or reject them. Rejected suggestions do not automatically become tracked brands. ## Historical recalculation Changing identity rules can affect old metrics. For example, adding an alias may reveal mentions in answers collected before the alias was known. Selfwin can rescan stored answer text after relevant brand changes. While recalculation is running, the brand records a recalculation state. Once complete, dashboards use the rebuilt mentions. ## Choosing competitors Track brands that could realistically be named in response to the same prompts. A long list of unrelated companies dilutes share of voice and makes position harder to interpret. A practical starting set is: 1. Your own brand. 2. Three to six direct alternatives. 3. One adjacent category leader if customers genuinely compare it. Use collected suggestions to discover unexpected competitors, then verify them against the project brief before accepting them. ## Removing a competitor Removing a competitor deletes its tracked identity and linked brand mentions from the project. It does not rewrite the original answer text. Your own brand cannot be removed from the competitor interface. --- # Brand insights > Compare a brand across models, topics, tags, countries, competitors, and source retrieval using trends and performance matrices. - Product: Selfwin - Section: Advanced analysis - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/insights - Keywords: brand insights, performance matrix, model comparison Brand Insights is a focused analytical view for one selected brand. It is designed to answer where the brand performs well, where it loses, and what changed around the same time. ## Strongest and weakest models The summary compares the selected brand’s performance across model channels. It identifies the strongest and weakest model for the active metric and filters. A model result with very little coverage should be interpreted cautiously. Check the answer count before treating it as representative. ## Trend analysis Choose one metric: - Visibility. - Share of voice. - Sentiment. - Position. The chart uses the selected date range and resolution. Position uses an inverted scale because a lower numerical rank is better. You can overlay retrieval for a selected domain. This helps test whether a visibility movement coincides with the model using your site more or less often. Markers show relevant prompt or model events recorded in the selected period. They provide context, not automatic causal attribution. ## Performance matrix The matrix crosses two dimensions and calculates the chosen metric for every populated cell. Available dimensions are: - Model. - Brand. - Topic. - Tag. - Country. Examples: - Model × Topic shows where one model behaves differently across subject areas. - Brand × Country compares competitors across markets. - Tag × Model compares commercial prompts with informational prompts by channel. The matrix includes the raw answer and mention counts needed to distinguish a strong result from a small sample. It can be copied as tab-separated text or exported as CSV. ## Model rankings The ranking table shows the leading brands for each model channel using the selected metric. Open the full Ranking page for a larger comparison. ## How to investigate a gap 1. Select your brand and the relevant date range. 2. Find the weak model or matrix cell. 3. Open the corresponding prompts and chats. 4. Inspect the sources and fanout queries behind those answers. 5. Run the [Visibility Agent](./visibility-agent) when the evidence represents a real opportunity. --- # Brand perception > Run an independent market study that measures how strongly AI answers associate each tracked brand with editable attributes. - Product: Selfwin - Section: Advanced analysis - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/perception - Keywords: brand perception, association, prominence, competitor radar Brand Perception is a separate analysis from ordinary tracking metrics. It asks model channels to evaluate how tracked brands relate to market attributes, then stores the result as an immutable run. ## Attributes An attribute is a concept you want to compare, such as: - Ease of use. - Reliability. - Enterprise readiness. - Value. - Customer support. Each attribute has a name and a set of matching phrases. Add attributes that describe real buying criteria in the project’s market. ## Run perception Select a model filter if needed, then choose **Run perception**. Selfwin performs independent model checks and stores: - Date range and answer count. - Attribute definitions used for the run. - Brand and model channel. - Prominence and association scores. - Supporting answer evidence. An old run does not silently change when new tracking answers arrive. Run perception again when you need an updated market view. ## Association Association measures how consistently the available evidence connects a brand with an attribute. ```text association = supporting overlap relative to the available brand and attribute evidence ``` A high association means the attribute is repeatedly connected to the brand when relevant evidence exists. ## Market prominence Prominence measures how strongly the brand occupies the attribute compared with tracked competitors. Earlier placement receives more weight; an absent brand receives zero for that check. Association and prominence can diverge. A brand may be strongly associated with reliability but still have lower market prominence if competitors are named earlier or more often. ## Reading the page - **Headline cards** summarize the strongest association, best competitive attribute, and largest gap. - **Radar chart** compares the shape of selected brands across attributes. - **Heatmap** shows prominence and association for every brand-attribute pair. - **Evidence list** provides the supporting answer and competitive rank. Use the evidence rather than treating the visualization as a standalone brand survey. The result reflects the model channels, competitors, attributes, and evidence included in that run. --- # Visibility Agent > Understand how Selfwin turns stored tracking evidence into classified gaps, proposed actions, and reviewed drafts. - Product: Selfwin - Section: Take action - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/visibility-agent - Keywords: visibility agent, GEO agent, content gaps, AI actions The Visibility Agent reviews recent tracking evidence and prepares work for prompts where your brand is weak. It is a closed-loop workflow: evidence becomes a gap, a gap becomes an action, and an action can later be measured against new tracking results. ## When a prompt is flagged The agent reviews active prompts over its analysis window. A prompt can be flagged when your visibility is below the configured threshold or when the leading competitor is sufficiently far ahead. It records a snapshot of the brand standings, URLs, citations, and fanout evidence used for the decision. ## Gap types ### Absent Your brand does not appear in enough answers for the prompt. The agent may propose a new page or another way to establish a relevant answer surface. ### Leak Your content or domain contributes to an answer, but another brand receives the visible credit. The evidence links the source visibility to the brand-visibility gap. ### Out-sourced Models rely on a third-party source where competitors are present and your brand is missing or weak. The agent may propose editorial, community, directory, or outreach work. ### Entity owned An owned page exists but does not win the expected retrieval or brand result. The agent may propose retargeting the page or fixing indexing. ## Action types Depending on the evidence, the agent can prepare: - Retarget an existing page. - Create a new page. - Produce a video brief. - Draft a community reply. - Draft outreach. - Create an indexing ticket. - Create a leak-resolution ticket. ## Drafting and linting On-domain drafts use the project brief, competitors, source URLs, and observed fanouts. They pass through Selfwin’s content linter before the run reports them as ready. Expensive research and drafting pipelines are budgeted per run. The agent can still create evidence-backed template actions when a provider is unavailable. ## Run behavior Runs can be started manually, on schedule, or through MCP. A completed scheduled run is idempotent for its project and week. Run history preserves status, configuration, summary, errors, and timestamps. The agent does not publish. Continue with [Action queue](./actions) to review and approve its output. --- # Action queue > Review, approve, dismiss, and measure the work proposed by Selfwin without allowing automatic external publishing. - Product: Selfwin - Section: Take action - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/actions - Keywords: action queue, approvals, GEO recommendations, drafts The Action queue is the human control point between Selfwin’s analysis and any external work. ## Two action sources Selfwin can surface: - **Visibility Agent actions:** Evidence-backed work tied to a specific agent run, prompt, and classified gap. - **Source recommendations:** Opportunities generated from recent citation patterns and competitor presence by source category. Both are recommendations. Neither is automatically published. ## Action states Agent actions move through: - **Draft:** Prepared and awaiting review. - **Approved:** A person accepted the proposed work. - **Shipped:** The work was completed outside or through an explicitly connected workflow. - **Dismissed:** The proposal will not be pursued. Source recommendations use open, done, and dismissed states. ## Review the evidence Before approving, check: - The prompt and current brand standing. - The classified gap. - Supporting URLs and source ownership. - Observed fanout queries. - Proposed target and action type. - Draft content and linter result. The score is a prioritization aid. It does not guarantee that a model will change its future answer. ## Approval is not publishing Approving an action changes its state inside Selfwin. It does not: - Edit your website. - Post to a community. - Email a publisher. - Submit an indexing request. - Send a message to a third party. External execution requires a separate, explicit workflow and the permissions appropriate to that destination. ## Measuring impact After work ships and new tracking evidence exists, Selfwin can store an attribution record with pre-action visibility, post-action visibility, the measured week, and a note. Attribution records a before-and-after observation. It should not be interpreted as proof that the action alone caused the change. --- # Slack, Claude, and MCP > Connect Selfwin to Slack or an MCP client while preserving project scope, revocable access, and approval boundaries. - Product: Selfwin - Section: Operate Selfwin - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/integrations - Keywords: Selfwin MCP, Claude integration, Slack AI employee, OAuth Selfwin can be operated from its dashboard, Slack, or an MCP-compatible client such as Claude or Codex. These surfaces share the same project data and approval rules. ## MCP access Selfwin exposes authenticated tools for project configuration, analytics, tracking, perception, sources, the Visibility Agent, and the action queue. Access can use a revocable Selfwin token or the supported OAuth flow. Tokens are shown only when created; Selfwin stores a hash rather than the original token. An MCP client can: - List and inspect projects. - Manage topics, tags, prompts, and competitors. - Read chats, fanouts, metrics, matrices, sources, and perception results. - Ingest an externally collected anonymous answer. - Start tracking or the Visibility Agent. - Read and update actions. Tool descriptions identify writes and destructive operations. Clients should still request human approval before irreversible or external work. ## Claude Claude can connect to Selfwin through MCP and OAuth. Once authorized, it can read the selected Selfwin project and operate the exposed tools within the user’s account. Claude integration does not mean Claude is automatically enabled as a tracking model. Tracking channels are configured separately in project Settings. ## Slack The Slack integration installs one project-scoped Selfwin employee in a workspace. It can provide briefs and answer supported questions using the same project evidence. The installation stores: - Slack team and bot identity. - Encrypted bot token. - Granted scopes. - Default channel. - Daily-brief preference and status. Slack event receipts make retried events idempotent. A repeated Slack delivery should not create repeated work. ## Security boundaries - MCP tokens are user-owned and revocable. - OAuth authorization codes expire and are stored by hash. - Slack installations are visible only to their owning authenticated user. - Server-only OAuth and Slack records are not exposed to anonymous or authenticated browser clients. - External publishing remains outside the default approval action. Open **Integrations** in Selfwin to create or revoke access and test the current connection. --- # Project settings > Configure project identity, model channels, collection method, schedule, time zone, brand rules, and integrations. - Product: Selfwin - Section: Operate Selfwin - Updated: 2026-09-06 - Canonical: https://docs.selfwin.co/settings - Keywords: Selfwin settings, model configuration, tracking schedule Settings controls the identity and collection behavior of the current project. ## Project profile The project profile contains the name, domain, description, and structured brief used by prompt generation and action drafting. Keep the profile factual and specific to the project’s actual product and market. ## Brand identity Configure the own brand’s display name, tracked name, aliases, domains, matching expression, and chart color. Changes that affect detection can trigger historical recalculation. See [Competitors](./competitors) for the same identity model applied to other brands. ## Model tracking Enable the model channels the project should run. Provider channels require their corresponding credentials. Browser channels require a working browser collector. The page reports whether each provider and browser method is currently configured. A label in the interface does not by itself mean the channel can run. ## Collection method Scheduled tracking can use browser or provider API collection. Choose based on the answer surface you intend to measure, then keep the method stable when interpreting trends. Browser and API answers are not interchangeable. Switching methods can change wording, sources, and fanout evidence even when the prompt stays the same. ## Schedule and time zone Choose daily or weekly answer collection and set a valid IANA time zone, such as `Europe/Berlin` or `America/New_York`. The project-local date determines daily deduplication and reporting windows. The weekly Visibility Agent and daily Slack brief are separate scheduled jobs. ## Run history Run history reports: - Trigger and collection method. - Planned, completed, skipped, and failed work. - Start, update, and completion times. - Error summaries. Use it to distinguish a true zero-visibility result from incomplete collection. ## Integrations and access Settings links to Slack, MCP tokens, and OAuth-connected clients. Revoke access that is no longer used. Rotating an external provider credential is handled by the environment or deployment where Selfwin runs. ## Project isolation Selfwin is configured to use its dedicated infrastructure. The application validates the Supabase project identity before running application migrations, which prevents a differently configured database from being modified accidentally.