AI search monitoring is a repeatable way to observe whether a brand appears, is mentioned, and is cited in answers to a stable set of buyer questions.
I built AI Search Vitals because traditional SEO reports were not answering a question I kept hearing:
When a potential customer asks an AI system for a recommendation, what does it actually say about our brand?
Traditional search gives you familiar signals: impressions, clicks, rankings, and backlinks.
The same buyer question can produce different results depending on the model, prompt wording, market, language, retrieval mode, source set, and time of day. A screenshot may show what happened once, but it does not explain whether the result is a pattern.
There are four problems I wanted to solve: A brand can be relevant to a category but never appear in the answer. A brand can be mentioned without being cited as a source. A competitor can be recommended while your better explanation is ignored. A single visibility score cannot explain what changed or what to fix next.
The hard part is not generating another score. The hard part is preserving enough evidence to understand the answer behind the score.
AI Search Vitals is an AI search visibility monitoring platform for brands, marketing teams, SEO and GEO practitioners, agencies, and founders.
The workflow starts with a project: Add a brand, website, market, language, and competitors. Save the buyer questions you want to monitor. Select the AI channels enabled for the workspace. Run an initial baseline. Review mentions, recommendations, competitors, and citations. Schedule daily or weekly observations. Use the results to decide which page, prompt, or content gap to improve.
The product keeps the evidence behind every successful observation: The original buyer prompt The actual model and provider Market and language Search mode and run timestamp Raw AI answer Brand and competitor mentions Ordered recommendation position when available Citation URLs, domains, titles, and snippets Token usage and request metadata
For example: Did the brand move from absent to mentioned? Was the mention accurate or misleading? Did the model cite our website or a competitor? Which page became the source? Did the change happen across a prompt group or only once?
AI Search Vitals also includes competitor share of voice, citation exploration, prompt research, monitoring history, CSV export, and GEO audit tools for crawlability, content readiness, and AI Query Expansion.
The application runs on Next.js 15, React, and TypeScript, with OpenNext and Cloudflare Workers handling the production runtime.
Workspace, project, prompt, run, observation, mention, and citation data are stored in Cloudflare D1 through Drizzle. OpenRouter adapters keep provider metadata, usage information, citations, and raw response context instead of reducing every response to a plain string.
Scheduled monitoring uses a server-controlled workflow: Cloudflare Cron finds prompts that are due. Each prompt and model target becomes a queue job. The worker executes the observation. The parser extracts mentions, positions, and citations. The observation is stored with an idempotency key. Retryable provider failures can be inspected without duplicating the result.
That is why AI Search Vitals stores prompt snapshots and encourages teams to separate discovery, evaluation, comparison, and task-based questions. A stable prompt set creates a baseline that can be reviewed over time. Mentions need context
The system needs to distinguish between a useful recommendation, a passing reference, an incorrect description, and a competitor comparison. A mention without context can lead to the wrong content decision. A mention is not a citation
It can also cite a page for one narrow fact without recommending the whole brand. These are different signals, so they need to be measured separately. Citation formats are inconsistent