Documentation & Guides

DoneHat generates buyer-intent questions people ask ChatGPT ("best project management tools for startups," "alternatives to [competitor]") and checks whether your product appears in the AI-generated answer. You get a clear pass/fail for each question along with a reliability score that rejects unsupported model-judge claims.

Enter your website URL on the brand-account setup screen. DoneHat crawls your site and builds a brand profile with your product names, value props, and category labels. Review the profile, add 2-5 competitor URLs, and click "Run scan." Scan time depends on question count, enabled platforms, provider availability, and queue load. Each answer check records whether your brand was supported by the answer evidence, what the AI or search surface said, and the extraction confidence.

Missing mention - the AI answered the buyer's question but never referenced your brand. This is the most common issue for new brands.

Weak mention - your brand appeared but in a subordinate position ("You might also consider...") or with incorrect information attached.

Competitor dominance - a competitor was named as the primary recommendation while your brand was absent or downplayed.

Inaccurate claim - the AI stated something factually wrong about your product, such as wrong pricing, deprecated features, or an incorrect founding date.

Missing context - the AI mentioned your brand but omitted critical differentiators like pricing tier, integration support, or target audience.

Open an issue and create an action or fix brief. DoneHat analyzes root causes such as missing structured data, thin content on key pages, weak brand facts, or weak third-party signals, then drafts a fix deliverable for human review. After you publish changes, run a proof check to measure the before/after difference.

AI models build answers from training data and retrieval-augmented sources. If a competitor has deeper product documentation, more third-party reviews, better structured data (FAQ schema, HowTo markup), and stronger entity relationships in knowledge graphs, the model treats them as the authoritative answer. Your content might rank well in Google but still be invisible to ChatGPT because the model weights different signals.

The competitor dominance issue appears when your competitor is named as the primary recommendation across multiple buyer questions while your brand is absent. DoneHat shows you the exact question, the competitor's position in the answer, and the sources the AI cited. You can see side-by-side comparisons of how each brand is described.

Entity alignment brief - DoneHat drafts a content plan that maps your brand attributes, features, pricing, and use cases to the question groups where you're losing. This ensures AI systems can match your content to the buyer's query.

FAQ structured data - add FAQ schema to your highest-traffic pages so AI retrieval systems can pull direct answers.

Comparison pages - create honest comparison content ("[Your product] vs [Competitor]") that gives models explicit signals about your positioning.

Third-party mentions - identify review sites, directories, and industry publications where your competitor appears but you don't. Each external citation is a stronger public proof point.

After implementing fixes, run a proof check from the action detail screen. DoneHat repeats the same questions that triggered the issue and compares the new responses against the original scan. You'll see whether your mention rate improved, whether you moved from absent to mentioned, and whether the competitor's dominance score decreased. Share the proof report with stakeholders as evidence of impact.

DoneHat launch scans start with ChatGPT answer observations. Additional answer sources such as Google, Claude, Gemini, and Perplexity are enabled only when provider credentials, plan limits, and runtime configuration support them. Each enabled answer source returns answer text or search evidence, source data when available, mention detection, competitor mentions, confidence, and failure coverage when a provider is unavailable.

A visibility scan takes your enabled buyer questions and checks them across every enabled answer source. If you have 20 questions and 3 answer sources enabled, that is up to 60 answer checks per scan. Each answer check records platform, question, answer text or failure reason, brand mention, competitor mentions, sources when available, sentiment, extraction confidence, and timestamp.

Snapshot: one lifetime public audit attempt with up to 6 sampled questions and 3 answer attempts per question, within the trial processing allowance. Includes answer evidence and verified gaps, without workspace agent drafts or proof checks. Founder Proof: 25 buyer questions, one fix brief, and one proof rerun. Solo: 25 buyer questions and weekly evidence reruns for one small brand. Scaler: 50 buyer questions, weekly reruns, and two fix briefs per month. Agency Lite: three client brands, 100 buyer questions, and client-ready summaries.

Monitoring reruns track changes over time. Proof checks compare results after a published fix under matched conditions.

The reliability score (0-100) summarizes supported visibility evidence across sampled answer sources and buyer questions. It factors in mention support, source evidence, answer consistency, and known failure coverage. Unsupported model-judge claims are rejected instead of counted as wins.

During brand-account setup, select your primary region from the dropdown (Bangladesh, Germany, United States, etc.). DoneHat uses this to generate region-aware questions like "best accounting software in Dhaka" or "top marketing agencies in Berlin" instead of generic global queries. This matters because AI models serve different answers based on geographic context.

DoneHat adapts question templates to include your region and market context. For example, "best CRM for small businesses" can become "best CRM for small businesses in Germany." This helps surface local competitors and localized claims that would be missed by generic global questions.

When you set a region, DoneHat's competitor discovery prioritizes brands active in your market. If you're a SaaS company based in Bangladesh, you'll see both global competitors and local alternatives that AI engines might recommend to buyers in your region. Add any missing competitors manually to ensure complete coverage.

Each observation includes region metadata so you can filter results by geographic context. Your reliability score reflects how well AI engines recommend you specifically to buyers in your region, not globally. This prevents inflated scores from markets where you don't compete.

DoneHat's assisted workflow follows the product loop: monitor issues from visibility scans, diagnose likely root causes, plan a fix strategy, draft deliverables, require owner or editor approval for sensitive actions, and repeat the same questions after fixes ship. The system handles analysis and drafting; your team controls what gets published.

FAQ page drafts - structured Q&A content targeting question groups where you are missing. Structured data snippets - JSON-LD recommendations such as FAQ, Organization, Product, or SoftwareApplication when applicable. Page rewrites - updated copy for existing pages with stronger entity alignment and claim clarity. Entity alignment briefs - strategic documents mapping your brand attributes to the questions where you need visibility. Comparison page outlines - content frameworks for "[You] vs [Competitor]" pages.

AI drafts are designed for human review. Owners and editors can approve, reject, and execute eligible actions. Viewer roles can inspect evidence but cannot make sensitive brand-account changes.

DoneHat uses configured brand facts as ground truth: one-liner, target customer, pricing model, differentiators, proof points, and forbidden claims. Fact checks and fix briefs use those fields to avoid unsupported claims and highlight places where AI answers misrepresent the product.

A proof check repeats the exact same questions that originally surfaced an issue, on the same platforms, and compares the new AI responses against the original scan. It's not a full monitoring scan - it's a targeted test of whether your specific fix moved the needle. Each proof check uses one proof credit from your plan allocation.

Improved - your brand now appears in the AI answer where it was previously absent, or your position improved, such as from footnote to primary recommendation.

Regressed - your visibility decreased compared to the original scan. This can happen if a competitor published stronger content or if the AI model was updated.

Unchanged - no measurable difference. The fix may need more time to propagate through the AI model's knowledge, or the root cause wasn't fully addressed.

Inconclusive - the AI response changed but in ways that don't clearly indicate improvement or regression. This sometimes happens with non-deterministic model outputs.

Do not expect every configured answer source to change immediately after you publish a fix. Some surfaces update from live retrieval, while others rely on slower indexes or model-side behavior. If the first proof check is unchanged, wait, re-check the cited sources, and decide whether the fix needs stronger page content, clearer schema, or more third-party evidence.

Every proof check stores a before/after comparison with timestamps, the questions tested, and quantified changes such as mention movement and positioning. Use reports, exports, or weekly digests to share proof with stakeholders.