How ModelSaid used its own platform to become #1 in AI visibility

ModelSaid became #1 in AI visibility by treating its own category presence as an operating system, not a launch campaign. The team established a prompt baseline, inspected the answers and sources behind the score, fixed the clearest evidence gaps, and repeated the same tests. That loop made improvement measurable and exposed regressions early. It is also an important qualification: "#1" is ModelSaid's brand claim based on its monitored category and prompt set, not a third-party award or a promise that every possible AI conversation ranks brands identically.

The starting point was a defined category, not a vanity query

A test such as "What is ModelSaid?" says whether a model recognizes a supplied name; it does not show whether a buyer would discover the company. ModelSaid instead built questions around real jobs: monitoring brand mentions in AI answers, finding inaccurate descriptions, comparing answer engines, and turning visibility gaps into work. Most discovery prompts did not contain the brand name. Each prompt had a clear intent and expected qualification, which prevented irrelevant mentions from being celebrated as wins. A team copying this approach should define its audience, market, use case, and meaningful competitors before running a baseline.

A baseline made the invisible problem inspectable

The first useful artifact was not a perfect score. It was a dated collection of complete answers across ChatGPT, Claude, Gemini and Perplexity, tied to stable questions. The team reviewed whether ModelSaid appeared, how it was described, what alternatives appeared, and which public sources supported the response. ModelSaid runs these observations through model APIs. An API response is not an identical copy of every consumer app screen, where account state, experiments, location and interface features can differ. It is a consistent measurement surface for comparing controlled runs.

Findings became small, attributable changes

• Unanswered buyer questions became direct, useful pages or clearer sections on existing pages.
• Ambiguous product language became explicit facts with scope and a maintained source of truth.
• Technical access issues were separated from content and reputation issues instead of being mixed into one score.
• Structured data was added only when it matched visible content and clarified a real entity or page type.
• Each material change was logged against the prompts it was expected to influence.

That last step matters. If ten unrelated edits ship together, a better answer teaches the team very little. ModelSaid favored bounded changes and then watched the relevant question group. The AI readiness assessment is a practical way to find foundational access and clarity problems before rewriting a whole site. The team also checked machine-readable markup with the schema validator, while remembering that valid schema can clarify evidence but cannot force a model to cite or recommend a company.

Repetition turned an improvement into category leadership

Generated answers vary, and models change. ModelSaid therefore kept a stable benchmark set, reran it on a schedule, and investigated meaningful movement at the answer level. New prompts were tested separately before joining the benchmark. This preserved a comparable core while allowing the measurement to follow new buyer language. A short-term mention could not carry the program; the brand had to remain accurately visible across relevant questions and supported by evidence. Over repeated loops, that discipline is what allowed ModelSaid to make and defend its #1 claim while staying near 100/100.

Copy the system, not the outcome

Another company should not copy ModelSaid's prompts or assume that near 100/100 is the right immediate target. Copy the method: define qualified discovery, save full answers, locate the evidence gap, assign one owner, make a bounded improvement, and retest the same questions. Report accuracy and recommendation context beside the aggregate score. Keep failed experiments in the log so they are not repeated as folklore. The free scan is useful for orientation; continuous monitoring is what reveals drift and whether a change persists. Compare the available cadence and workflow features on ModelSaid pricing, then choose the smallest plan that matches how often your team can genuinely review and act. Measurement earns trust through visible boundaries.