How we measure, and where measuring has limits

Every answer we collect comes from the engine's official API. ChatGPT, on OpenAI's API, is live today. Claude on Anthropic, Gemini on Google, Perplexity's Sonar and Grok on xAI are built to the same standard and rolling out next. Where the API supports web search or grounding, we turn it on, because that is the mode buyers actually experience.

We do not scrape consumer interfaces. That is why Microsoft Copilot and Google AI Overviews are absent from the product: neither offers an official API today. When they do, we will add them. We would rather show you an engine honestly than through a side door that breaks silently.

API answers are a close proxy for the consumer products, and that is the honest phrase for it: a proxy. Ranking sources, retrieval and model versions match closely, but the consumer apps add personalization we cannot see. Every vendor in this market has the same gap. We are the ones telling you about it.

Onboarding drafts your question set from your own site: best-in-category asks, alternatives to named rivals, use-case questions, and head-to-head comparisons. You edit and approve every one before anything runs.

Questions that contain your own brand name are handled separately, and this is the detail most tools get wrong. If a question names your brand, the answer will mention you by construction, and counting that as visibility inflates the score. So branded questions never feed your headline number, and they only count as won when the engine actively picks you, which is the thing that actually matters in a comparison.

Each answer is parsed for brand mentions, and every claim must survive verification against the raw text. A mention counts only if the brand's name literally appears in the answer. A recommendation counts only if the exact recommending phrase exists in the answer, verbatim. Claims that fail verification are discarded.

This strictness exists because language models parse the answers, and unaudited model output drifts optimistic. We built the audit into the pipeline, and we keep the full answer text so you can open any question and check our scoring against the source yourself. If you ever find a mismatch, we want to hear about it.

The visibility score is the share of discovery answers, latest answer per question per engine, that mention your brand. When you recheck, the newest answer replaces the old one in the score, so a fix that flips an answer shows up immediately. History is never rewritten: the trend chart keeps a point for every completed check.

Partial checks are safe by design. If you stop a check midway or check only newly added questions, the untouched questions simply keep their previous answer. A partial run can nudge your score; it can never fake a collapse or a spike.

The same question asked twice can produce different answers; SparkToro measured under 1% chance that two identical ChatGPT runs return the same brand list. That variance is a property of the engines, and pretending otherwise would be selling you false precision.

Our defaults are built for it: per-check snapshots rather than single-moment claims, trend lines over one-off readings, and the raw answers always retained. Treat any single check as one sample of a moving target, and treat the trend as the truth. Higher-frequency sampling for tighter confidence is on the roadmap for Growth and Agency plans.

Generated fixes follow the published evidence on what earns citations: answer-first structure, concrete statistics with named sources, and cited references, patterns measured by the Princeton GEO study (KDD 2024) at roughly 30 to 41% citation lift, and consistent with large citation analyses from Ahrefs and Semrush showing engines lean heavily on third-party content.

Each fix is researched on the live web at generation time, and every number in it comes from that research, never from the model's memory. You review before publishing, always. The fix also lists the third-party sites the engine actually cited for your question, because being named on a source AI already trusts often moves an answer faster than your own page.

No scraped engines. No paid backlink schemes. No Reddit or forum astroturfing, which violates platform rules and eventually burns the brands that fund it. No auto-publishing to your site without your review. And no unlimited plans that quietly throttle you; usage is metered in credits and every cost is shown before you click.