A careful snapshot, not a promise.

What we test

Each dental audit uses 10 or 11 discovery prompts that do not include the practice name, plus one recognition control that does. The discovery set covers local discovery, public access information, trust signals, one existing service selected by the practice, and urgent care only when the practice confirms it currently offers urgent or emergency appointments. Every prompt runs twice on each included engine. One response means the text generated by one engine for one prompt in one run.

Why this is not a Google ranking report

Traditional search commonly presents ranked links and map results. AI systems can synthesize websites, public profiles, reviews, directories, and other sources into a shorter generated answer. A practice can therefore rank visibly in search and still be omitted from an AI response—or appear in one AI system and not another. Seenward measures observed recommendation frequency, citations, prominence, and repeat-run stability; it does not convert Google rank into an assumed AI result.

How the Model Standing Score works

Each engine receives its own score using only its 20 or 22 eligible discovery responses. Recognition responses are preserved as evidence but excluded. The score combines how often the practice was named (55 points), how often its website was cited (15 points), how often it appeared among the first three options (15 points), and visibility across eligible question categories (15 points). When multiple engines complete, the overall score is the rounded mean of their separate scores. A score is withheld when too few responses complete.

The 55/15/15/15 weights are Seenward’s current judgment-based rubric. They emphasize discovery while preserving credit for evidence, prominence, and breadth. They have not been validated against patient demand, revenue, or conversion outcomes.

How to read the target

The internal exploratory reference threshold is 50 or higher. It is not a statistically validated industry benchmark, average, or promised outcome. The limited August 2026 calibration contains 10 practices, with a median of 14.5 and a highest observed score of 60. We will publish stronger distribution information only when the cohort supports it.

Why we repeat every question

AI responses are probabilistic. Seenward separates questions that are stable-visible, variable, or stable-absent across two runs. Repetition does not eliminate uncertainty, but it is more honest than treating one response as a permanent fact. The report preserves the exact question, run number, response, mention result, and cited sources.

The new-patient discovery path

The report translates the saved AI responses into four public-information stages: whether the practice was discovered, supported by a first-party citation, placed prominently for comparison, and surfaced for access-intent questions. This is a visibility diagnostic—not a conversion funnel, demand estimate, or assessment of practice operations.

Local response context without duplicate runs

Seenward identifies other practices incidentally named in the same eligible discovery responses. This avoids separate competitor-query costs, but it is not a predefined or geographically validated competitor cohort and is not adjusted for proximity, market share, or search volume. Multiple practices may appear in one response. The names are reviewed before delivery and presented as co-mention context—not a market, quality, reputation, or business-performance ranking.

Technical and public evidence

The technical review checks whether the business website can be reached and whether identity, services, structured information, and other useful signals are clear. The public-footprint review looks for independent or official sources that can corroborate the practice. A source being found does not mean every claim on it is verified.

What the audit cannot prove

The benchmark measures only the provider APIs and model IDs marked Tested in the report. Seenward supports OpenAI ChatGPT, Anthropic Claude, Google Gemini, and xAI Grok, but it never infers a result for an engine that did not complete. API results do not reproduce a free consumer chat account, personalization, private history, Copilot, Perplexity, or another untested product. Results can vary between runs and change as models and public information change. Repetition reduces the risk of overreacting to one response, but it does not remove uncertainty or establish causation.

Evidence and reproducibility

The private browser report preserves a response ID, provider, model, exact prompt, run number, unmodified answer, mention decision, cited sources, and source-review status. The report separately records the test timestamp, evidence retrieval date, report preparation date, and human-review status. The PDF summarizes that register and directs the client back to the secure browser evidence rather than shrinking dozens of raw responses into unreadable pages.

Human review

A completed report is held for human review before it is approved for delivery. We check co-mentioned practice names, identity mismatches, claims needing source verification, failed responses, and proposed visibility corrections that overstate the evidence or drift into clinical or business advice. A missing citation alone does not prove an AI statement is false, so Seenward does not report a hallucination rate without claim-level verification.

Scope and compliance boundary

The audit uses public business information only. It does not request patient information, assess treatment, certify HIPAA or legal compliance, or recommend changes to a practice’s clinical services. Any generated implementation material must be reviewed by the dentist and web developer before publication.