# Gumshoe: Frequently Asked Questions Answers to common questions about using Gumshoe, the AI search visibility platform. Last reviewed September 2026. For the platform reference covering metrics, model coverage, and methodology, see . --- ## Getting started ### What is Gumshoe? Gumshoe measures how AI models describe your brand when buyers ask for recommendations in your category, and gives you the diagnostics and content tooling to change those answers. It runs buyer-persona conversations against 11 model families from seven providers, records every answer, and reports where your brand appears, where competitors appear instead, which sources the models drew on, and how they characterize you. ### How do I start? Sign up at app.gumshoe.ai/go and run a free sample audit. No credit card is needed. Results come back within minutes. A work email is recommended, because it unlocks the standard account role and its entitlements. A personal address still works and still gets the free sample. Disposable, throwaway domains are the only ones turned away. ### How many free reports do I get? One free sample audit, per customer and per user, with no expiry. The sample is a Visibility Audit at reduced scope: 4 personas, 5 prompts per persona, across 4 models, for 80 conversations. It cannot be rerun or scheduled, but it can be upgraded to a paid plan at any time, and the personas and prompts a full audit would have used are visible in the sample as a preview. ### What is a project? A project is a brand plus a focus area, usually one product or service line. Plans are bought per project. A company tracking three product lines runs three projects, and they can sit on different plans. Agencies run one project per client brand. There are no per-seat charges, so anyone invited to a project can see it. ### How long does a report take to run? Most complete within a few minutes. Reports with many personas, prompts, or models take longer. You get a notification when results are ready, and the report shows progress while it runs. Runs involve hundreds of calls to several model providers, some of which retrieve live web pages before answering, so wall-clock time depends partly on provider response times. --- ## Building a report ### What is a Focus and why does it matter? The Focus is the product or service the report is about. It is the single most consequential setting, because it drives the personas, which drive the topics, which drive the prompts. Everything the report measures descends from it. Set a Focus that matches how buyers actually shop. "Project management software for construction teams" produces prompts a brand can win or lose on specific merits. "Software" does not. If a brand has several distinct lines, run a separate project per line rather than one report trying to cover all of them. ### How do I choose good topics? Topics are the themes that drive prompts. Think of them as buckets of traditional SEO keywords. Gumshoe generates topics automatically, but only you know your industry, so review and edit them. Once the topics and personas are right, the prompts follow without needing to review each one. What makes a topic good: - It is a genuine subcategory of the report's Focus, not a neighboring category. - It reflects products, services, or experiences you actually offer. - It uses the language buyers use, not internal jargon. "Tools to manage customer data" beats "MDM solutions". - The set spans the journey, from early research through purchase to post-purchase. - The set mixes broad categories with specific long-tail opportunities. ### Why do personas matter more than prompt wording? Because models answer the same question differently depending on who appears to be asking. A procurement lead, a practitioner, and an agency buyer get different recommendations for the same category, and a brand can be strong with one and invisible to another. Persona segmentation is what turns a visibility number into something you can act on. A brand at 62% overall that is at 85% with its core buyer and 30% with the segment it wants to expand into knows exactly which gap to work. Spend your review time on the Focus and the personas. The prompts will fall into place. ### How many personas and prompts should I use? The monthly Visibility Audit default is 8 personas with 10 prompts each, which is the configuration the included conversation allowance is sized for. More personas widen the range of buyers you can see, and six is a reasonable floor for a brand with a mixed audience. More prompts per persona give each persona a more stable number, and ten or more keeps a single unusual answer from moving that persona's percentage much. Custom reports go up to 100 personas and 50 prompts per persona. ### Can I edit or import personas, topics, and prompts? Yes on paid plans. Personas, topics, competitors, and prompts are all editable, and personas can be imported from a previous report so a consistent set can be reused across projects. The free sample is read-only. It shows the personas and prompts a paid plan would run, greyed out, so you can see what the fuller scope covers. ### What does regenerating prompts do? It replaces the current prompts for a persona or a report with a freshly generated set, based on the current Focus, topics, and personas. Two things to know. Past runs are not affected: completed runs keep the prompts they ran, so history stays readable. And future runs measure a different question set, so the trend line has a discontinuity at that point. Treat a regeneration as the start of a new baseline rather than a continuation of the old series. Regenerate when the Focus or topics have changed materially, when the prompts do not sound like questions your buyers would ask, or when the report is measuring the wrong category. Do not regenerate routinely, because the stable question set is what makes the series comparable. ### Can I save a report and come back to it? Yes. A report can be built, configured, and revised over as many sessions as you like. Nothing is charged until a run is confirmed, and the conversation count is shown before you confirm. ### Can I run a report in another language or for another country? Yes. A report can be configured with a language and a country. Language applies to prompt generation and to the requests, so a report configured in German asks German questions and reads German answers. Geography is applied where the channel supports it. Google AI Overviews takes a genuinely geo-located request, which makes it the most reliable signal for market differences. Other channels receive location as context rather than as routing, so read cross-market comparison on those as directional. --- ## Running and scheduling ### What does a subscription actually run? On both paid plans, a full Visibility Audit runs monthly and a Snapshot Audit runs on a shorter cycle: weekly on Starter, daily on Pro. The Snapshot Audit is a smaller sample of the same measurement, at 4 personas, 5 prompts each, across 4 models, for 80 conversations. Its job is to show that something moved while there is still time to act. The monthly audit is what tells you what moved and for whom. ### Why should I run reports on a schedule instead of once? Because a single run is a snapshot of a moving target, and almost everything that changes your visibility changes without you doing anything. - Model providers ship updates, and a provider's update can move a brand's position on its own. - Model families that use web search retrieve live pages at answer time, so new third-party coverage reaches answers immediately. - Competitors publish, earn placements, and improve their own technical readability. A one-off report tells you where you stand today. A schedule tells you the direction, which is the thing you can actually manage against. It is also what lets you attribute a change to work you shipped rather than guessing. ### Should I rerun a report after changing my content? Yes, but give the change time to propagate. Web-search models can pick up a changed page quickly, sometimes within a run or two. Pretrained model behavior moves much more slowly, and third-party sources have to publish before they can be cited. A rerun the same afternoon mostly measures noise. A practical approach is to let the scheduled cadence catch it, and read the change against the series rather than against a single prior run. ### Why do results vary between runs? Partly genuine change, partly sampling. Genuine change comes from model updates, live retrieval returning different pages, and competitor and third-party content moving. Sampling variance is inherent to measuring a percentage over a finite set of conversations, and models are non-deterministic by design. Gumshoe's answer to this is breadth rather than repetition: many distinct prompts asked once each across multiple personas and models, rather than the same prompt asked repeatedly. That produces a number that describes a population of buyer questions and is more stable run to run. Read small movements against the trend, not in isolation. Read large movements by opening the conversations and the Leaderboard to see what actually changed. ### How do I change which models a report uses? Model selection is a report setting on paid plans. Changing it affects future runs; past runs keep the models they ran. Both paid plans run six model families by default. Starter can change the selection within the available families. Pro draws from its own pool and has no cap on how many families a single audit runs, though each additional family multiplies the conversation count. Be aware that changing the model set changes what the trend line measures, for the same reason that regenerating prompts does. --- ## Reading your results ### What are the main numbers in a Gumshoe report? | Metric | What it means | | --- | --- | | Brand Visibility | Percentage of the report's answers that mention your brand | | Topic Visibility | The same percentage, restricted to one topic | | Persona Visibility | The same percentage, restricted to one persona | | Model Visibility | The same percentage, restricted to one model family | | Mentions | Raw count of times a brand is named across answers | | Rank | The position a model gives a brand inside a ranked answer | The breakdowns all use the same calculation as the headline number, which is what makes the headline decomposable. A brand at 62% overall that is at 90% on one model and 20% on another has a channel problem, and the model breakdown is what reveals it. ### What is the difference between visibility and rank? Visibility is how often you are mentioned. Rank is how strongly you are recommended when you are mentioned. They move independently. A brand at 90% visibility that consistently sits sixth in ranked answers is recognized but not led with. That is a positioning and authority problem, and it is invisible if you only look at the visibility percentage. Rank is on the Conversations page, filterable by model. ### What is the Leaderboard? A ranking of every brand in the report by mentions and visibility, measured across the identical set of prompts, personas, and models. Because all brands share one question set, the comparison is controlled. A competitor's gain and your loss are the same event seen from two sides, rather than two separately configured runs being compared. A report tracks up to 30 competitors. ### How do I read the heatmaps? Heatmaps cross two dimensions, such as persona against competitor, topic against competitor, or model against competitor. Each cell can show visibility as a percentage, mentions as a count, or the number of answers the cell is based on. Read the answer count first. A cell with a striking percentage over a small number of answers is a weaker signal than a modest percentage over many, and the count is what tells you which is which. Then read across rather than at individual cells. A row that is uniformly weak is a persona or topic problem. A column that is uniformly strong for one competitor is a competitive position. A single dark cell in an otherwise even row is usually worth opening the conversations for. ### What is the Conversations page for? It holds every prompt-answer pair in the run, filterable by persona, topic, model, and brand, with the full answer text and the sources cited. It is where you go when an aggregate looks wrong or surprising. Aggregates say what happened; the conversations say why, in the model's own words. It is also the fastest way to check whether a low number is a real absence or a misconfigured report. ### What do the Sources and Sources by Category sections show? The domains models cited when answering, how often each appeared, and which brands those sources support. Sources by Category groups them, so you can see whether a category leans on review sites, editorial coverage, community forums, or vendor documentation. This is the practical input to a digital PR plan. Knowing you are invisible is a problem statement. Knowing that a handful of domains account for most of the category's citations, and that a competitor appears on most of them, is a target list. ### Why do models cite pages that do not mention the brand they recommend? Because retrieval and recommendation are separate steps. The model retrieves pages to ground itself on the category, then produces a recommendation drawing on both those pages and its training. A citation is evidence of what the model consulted, not proof the cited page endorsed the brand. That is why sources are worth reading alongside the brands the answer named, rather than treating a citation as an endorsement. ### Does Gumshoe show real-world prompt volume? No, and no tool can. Model providers do not publish query volume. Any product showing "prompt volume" for AI search is estimating it from traditional search keyword data. Gumshoe measures what models say, not how many people asked. The prompts are representative buyer questions generated from your Focus and personas, not a sample of real traffic. What you can measure: your share of answers across a defined question set, how that compares with competitors on the identical set, which personas and topics you are weak on, which sources the category relies on, and how all of it moves over time. ### How does visibility relate to referral traffic from AI tools? They measure different things and routinely disagree, in both directions. High visibility with low referral traffic usually means the answers resolved the question, so there was no reason to click. That is a successful recommendation that produced no session. Referral traffic with low visibility usually means the traffic is arriving from informational queries while the report measures evaluation-stage recommendation questions. Both are real; they are just not the same question. When the two look contradictory, check which landing pages the AI referral traffic reaches and whether those pages match the topics and personas in the report. If they do not, you are comparing two different measurements rather than finding an error. ### Why is my visibility low, or zero? Work through it in this order. 1. **Read the conversations.** If the answers are about a different category than you intended, the Focus is wrong. Fix that first, because everything else descends from it. 2. **Check for a brand-name problem.** If answers name a variant of your name, or describe a different company with a similar name, it is attribution rather than absence. Brand name, URL, and summary are editable on the report. 3. **Check the Leaderboard.** If real competitors are being recommended and you are not, the number is correct and the question becomes why. 4. **Check the sources.** If the domains the models rely on do not cover you, that is the gap, and it is a digital PR problem rather than an on-site one. 5. **Check persona and topic coverage.** Weakness concentrated in specific rows is more tractable than weakness spread evenly. 6. **Check the Technical Audit.** If models cannot fetch or parse your pages, nothing else you do on-site will land. Strong traditional search performance alongside zero AI visibility is not a contradiction. Models draw on a different mix of sources and reward different properties. --- ## The five audits ### What audits does Gumshoe run? | Audit | What it answers | Availability | | --- | --- | --- | | Visibility Audit | Where you appear across models and personas, and who appears instead | All plans | | Snapshot Audit | Whether anything moved since the last full audit | Starter weekly, Pro daily | | Sentiment Audit | How models talk about you, not just whether they mention you | Pro | | Citation Audit | Which sources models lean on, and who those sources support | Pro | | Content Audit | What your site is missing against the questions buyers ask | Pro | | Technical Audit | Whether a model can fetch, read, and extract an answer from a URL | 1 run free, 10/period Starter, 50/period Pro | ### Does Gumshoe track sentiment? Yes. The Sentiment Audit is included on Pro. It generates questions at two buyer-journey stages, Product and Vendor Aware, and Evaluation and Decision, most of them naming competitors directly, because a model scores what the question puts in front of it. Answers are read several times by a judge that extracts themes per brand. Themes are clustered, scored for relevance to your Focus, and presented as pros, cons, and caveats, with a topic scatter and a heatmap. The output is deliberately thematic rather than a single score. "Models consistently raise implementation time as a caveat" is something a content team can act on; a sentiment value of -0.2 is not. ### What does the Citation Audit tell me? Which domains models actually cite in your category, how often, and which brands each source's coverage supports, separating your own domain from third parties. Citation data comes from the model families that run with web search on, since those are the ones that report the sources they used. ### What does the Content Audit tell me? Where your site fails to answer the questions buyers are asking, split into two distinct failures. A **retrieval issue** means the right content is missing, weakly matched, or not being surfaced. A **generation issue** means relevant content exists but does not give the model what it needs to build an answer from it. A third state means coverage is strong and the job is to keep it current. The distinction matters because the fixes are opposite. Retrieval issues usually need a new page or a structural change. Generation issues usually need an existing page rewritten so the answer is stated plainly and early rather than buried or hedged. ### What does the Technical Audit check? It fetches a page and evaluates it against roughly twenty deterministic checks, then rolls the results into a score from 0 to 100 with per-category breakdowns. Coverage spans structured data (JSON-LD presence and validity, article schema authorship), document structure (heading hierarchy, main content landmark, content volume, content chunking), metadata (title quality, canonical tag, Open Graph, Twitter card, published date, visible byline), media and accessibility (alt text coverage, media captions), and crawlability (HTTP status, robots directives, mobile viewport, text-to-HTML ratio). Checks are applicability-gated, so a page is only scored against the ones that apply to it, and each result carries the evidence that produced it rather than a bare pass or fail. Runs are metered: 1 on the free sample, 10 per billing period on Starter, 50 on Pro. ### How do I improve my AI visibility? There are three levers, and they act on different parts of the pipeline. 1. **Technical readability.** Whether a model can fetch your page, parse it, and extract an answer. The Technical Audit measures this, and it is the cheapest lever because the fixes are concrete and mostly one-time. 2. **Your own content.** Whether you have answered the questions your buyers ask, in a form a model can lift. The Content Audit measures this, separating missing content from unusable content. 3. **Third-party citations.** Whether the sources models rely on in your category cover you. The Citation Audit measures this, and it is the slowest lever but often the one that moves the most, because it changes what the models see rather than what you publish. Most brands are weakest on the third and start with the first. Working them in that order is reasonable, as long as the third one starts early, because it takes the longest to pay off. --- ## Content generation ### What is content generation? Gumshoe identifies the persona and topic combinations where your brand is weakest, then generates structured content aimed at that specific gap. A **piece** is one persona-topic combination, and it is the unit both the allowance and the billing count. A **run** is one execution that can produce several pieces. Starter includes 3 pieces per billing period and Pro includes 10. Further pieces are $25 each. Allowances reset on your renewal date, not the first of the month. Generated content exports to DOCX. ### Will generated content hurt my visibility? The concern behind this question is real: publishing undifferentiated filler at volume is a known way to damage a domain. Gumshoe's content is generated against a specific gap found in your own report data, which is what separates it from volume content. Each piece targets a question a defined persona actually asks about a topic you are measurably weak on, and the report says why you are weak on it. Generated content is still a draft, not a publication. It should be reviewed, given your own examples and evidence, checked for accuracy, and edited into your voice before it goes out. The generator gets you a well-targeted starting point; the differentiation is still yours to add. ### Where should I publish it? On your own domain, in the part of the site that matches the buyer stage the piece targets, and where your site's internal linking will actually reach it. An evaluation-stage comparison piece buried three levels under a blog archive is a retrieval problem waiting to happen. The Technical Audit is worth running on the published page afterwards, because a good piece that a model cannot parse does not help. --- ## Plans and billing ### What does Gumshoe cost? Two monthly plans, priced per project. | | Free sample | Starter, $99/mo | Pro, $299/mo | | --- | --- | --- | --- | | Visibility Audit | One time | Monthly | Monthly | | Snapshot Audit | No | Weekly | Daily | | Sentiment Audit | No | No | Included | | Citation Audit | No | No | Included | | Content Audit | No | No | Included | | Technical Audit | 1 run | 10 runs | 50 runs | | Content generation | No | 3 pieces | 10 pieces | | Included conversations | 80 | 480 | 480 | | CSV and JSON export | No | Yes | Yes | No annual contracts, no per-seat charges, cancel anytime. For a high volume of brands, talk to sales. ### What is a conversation, and what happens if I go over? A conversation is one prompt sent to one model, asked as one persona, plus the answer. Conversations are personas multiplied by prompts per persona multiplied by models, which is why the standard audit at 8 x 10 x 6 comes to 480. Both paid plans include 480 conversations per billing period. Beyond that, conversations bill at $0.10 each. The conversation count is always shown before you confirm a run. ### When am I charged? Subscriptions renew monthly per project. Scheduled runs charge when the run completes. Overage bills on what was actually run. Building a report costs nothing. You can configure and revise a report indefinitely; the charge attaches to a run, not to a draft. ### What is free? The sample audit, one per customer and per user, with no card and no expiry. Building and configuring reports. Viewing every report you have ever run, including after a subscription ends. ### How do I stop being charged? Cancel or pause the subscription on the project. Pausing keeps the report configuration and history intact and stops the scheduled runs. Your past reports remain viewable either way. ### What happens if a payment fails? You are notified, and the subscription moves to a past-due state. Updating the payment method resolves it. Scheduled runs do not proceed while a subscription is not in good standing, so the practical cost of leaving it is a gap in your trend line. ### Can I get a refund? Refund requests go to support@gumshoe.ai. Because runs consume real model provider capacity at the moment they execute, refunds are handled case by case rather than by a blanket policy. --- ## Data, exports, and integrations ### How do I export a report? CSV and JSON export are included on every paid plan. CSV suits spreadsheet analysis and reporting. JSON carries the structured data, including visibility scores, mention counts, personas, topics, prompts, models, sources, and full answer text, which is the format to use for warehousing or building your own dashboards. Generated content exports to DOCX. ### Does Gumshoe have an API? Yes. There is a documented public REST API versioned at `/v1`, described in OpenAPI. It covers creating and initializing reports, triggering runs, listing runs and checking run status, retrieving a run by ordinal, retrieving raw run data, and listing and retrieving page audits, at both organization and report level. Authentication is by bearer token. An organization admin enables API access and manages keys. ### Can I connect Gumshoe to Claude or another AI client? Yes. Gumshoe runs a Model Context Protocol server, so an MCP-capable client can query your audits directly rather than working from an export. Available tools cover listing your Visibility Audits and reading a given audit's models, personas, topics, prompts, brands, and citations, including cited URLs and answer text where the operation requires it. Authentication is OAuth against your Gumshoe account, and listing is scoped to the workspaces you are a member of. ### How do I build a reliable trend line? Key on the report, the run ordinal, and the completion timestamp. Every run carries these, so runs of one report form an ordered series. The thing that breaks a series is configuration drift. Changing the Focus, regenerating prompts, editing personas, or changing the model selection changes what is being measured, and a step change afterwards is an artifact rather than a result. Past runs keep their own configuration, so the break is visible rather than retroactive. Treat any configuration change as the start of a new baseline and record where it happened. --- ## Accuracy, methodology, and trust ### Why does Gumshoe use APIs instead of scraping? Every model is reached through an official provider API. Two reasons. Scraping breaks most providers' terms of service, which makes it a poor foundation for a measurement product a business is going to rely on. More importantly, scraped data is not the measurement you want. A scraped session carries personalization, caching, session state, and interface artifacts, so what you get is one anonymous browser's view of a model rather than the model. An API call is a controlled request with a known model and a reproducible configuration. ### How do you keep results unbiased and consistent? Every brand in a report is measured against the identical set of prompts, personas, and models, including your own. Prompts are submitted once each rather than resampled, so no question is weighted more heavily than another by accident. Answers are parsed for every brand present, not only the report's own brand, which is how competitors end up in the Leaderboard without being configured favorably or unfavorably. The check on all of it is that the underlying conversations are readable. Every aggregate can be traced to the answers it came from. ### Does persona optimization help if the user is not signed in? Yes, though less. For a signed-out user, models still use signals available in the session, such as language, location, and defaults, to shape an answer, so persona-aligned content can still influence tone and framing within that response. The larger payoff is with signed-in users, whose preferences and history persist across sessions, letting a model connect a brand to a need more consistently over time. ### What is Gumshoe's security and data posture? Gumshoe is data-minimal by design. The inputs are brand names, public URLs, and persona and topic selections. The outputs are model responses to those inputs. - No customer records, credentials, financial data, or private systems are ingested or accessed. - Personas and prompts are not shared outside your account or used for data harvesting, beyond being sent to model providers to generate your reports. - Model interactions run over official, approved provider APIs. - Internal access is role-based. The database has no public IP and is TLS-only, administrative access runs through an identity-aware proxy, secrets sit in a managed secret store with per-service access, and the database is regionally redundant with point-in-time recovery. Because the platform works with public information and model responses, many enterprise requirements that attach to sensitive data do not apply here. If a vendor review needs specifics beyond this summary, contact the team rather than inferring a certification from it. ### Does Gumshoe have Terms of Service and a Privacy Policy? Yes. Terms of Service: . Privacy Policy: . --- ## Troubleshooting ### My report is stuck or taking too long A run makes hundreds of calls across several providers, some retrieving live pages, so a large report legitimately takes longer than a small one. Provider slowness or rate limiting extends it further. If a run has been going far longer than previous runs of the same report, contact support@gumshoe.ai with the report and run details rather than starting a second run, since a duplicate run consumes conversations. ### My brand is misidentified or the name is wrong Signs of this: answers describe a different company, the brand appears under a variant spelling, or mentions attach to an unrelated business with a similar name. Brand name, URL, and summary are editable on the report, and correcting them usually resolves it on the next run. Because every conversation is stored whole, you can confirm the diagnosis by reading the answers rather than inferring it from the aggregate. For cases that persist, contact support@gumshoe.ai. ### I cannot edit my topics, personas, or prompts Free sample reports are read-only by design. They show the personas and prompts a paid plan would run, greyed out, as a preview of the fuller scope. Editing unlocks on a paid plan. Some capabilities also depend on your role in the workspace. If you are on a paid plan and still cannot edit, check your role with a workspace admin. ### Can I change the location on an existing report? Location is part of a report's configuration. Changing it changes what future runs measure, so treat the change as a new baseline rather than a continuation of the existing trend line. Past runs keep the location they ran under. --- ## Contact - Product: - Application: - Help Center: - Support: support@gumshoe.ai