# Product and measurement guide

## The practical workflow

| Step | What the team does | What it can learn |
|---|---|---|
| Set the question | Choose a brand, relevant competitors and buyer questions | Which specific discovery situations the investigation covers |
| Observe | Collect sampled responses through supported provider/model routes | What those runs returned in the recorded context |
| Inspect | Read answers, brand references, source citations and competitor context | Where an answer gets information and how a brand is represented |
| Diagnose | Review website accessibility, information clarity and content gaps | Candidate improvements to pages and supporting content |
| Review and act | Check generated proposals and use an available delivery path | What will change, where, and under whose approval |
| Re-measure | Compare later observations with their sample conditions | Whether the observed pattern changed; causality requires more evidence |

## Capabilities and boundaries

**Monitoring.** Track chosen prompts and investigate answers over time. Sampling coverage and frequency depend on the account, plan and supported routes. Brand discovery, comparison and branded queries answer different questions and should be described separately.

**Source inspection.** Inspect citations and the domains supporting an answer. A frequently cited source is not automatically accurate, independent, endorsed or causally responsible for a recommendation.

**Website diagnostics.** A crawl can reveal access, structure and information issues in the pages it reaches. An unsuccessful crawl is an observed retrieval problem; it does not establish that every AI service is unable to access a site. Diagnostic criteria are not universal ranking rules.

**Recommendations and content.** Findings can inform proposals for website or content changes. Generated claims and code need review against the actual business and site. An automatically generated suggestion is not proof of factual correctness or expected commercial impact.

**Delivery.** The public product describes connected website/CMS workflows, GitHub review paths and manually applying exported changes. Availability and action type depend on the integration and permissions. A message/task destination is not equivalent to publishing a web page. Verify the particular path before demonstrating it.

**Agency work.** Agencies can organize work around client brands and workspaces. Do not infer that a whole portfolio is one comparable sample; record the brand, workspace and question set behind each result.

**Free diagnostic.** The AEO Readiness Check is a public entry point for website investigation. It is not a live survey of all consumer AI applications, and a high score does not guarantee that an answer will mention or cite a business.

## What the numbers mean

The public [How we measure](https://omniimpact.net/docs/methodology/how-we-measure) guide, updated October 11, 2026, lists the engines on each plan, how often each checks and what each dashboard number means. It replaced the earlier in-app Methodology page. A deeper walkthrough of the method is available from the company on request.

| Term | Plain-language meaning | Reporting caution |
|---|---|---|
| Mention rate | The share of recorded answers that name the brand | Name detection and the counted answers matter |
| Citation rate | The share of recorded answers that cite the brand's site or the brand | A link is not the same as a positive recommendation |
| Share of voice | The brand's share of all brand mentions in answers to category questions that do not name it, including brands it does not track; a recommendation counts for more than a passing mention | Changing the question set or counting rule changes the result |
| Scores | Indexes from 0 to 100, such as the Website, Prompt and Overall scores; not percentages of answers | A rate and a score can move differently |
| Recommendation | An answer actually selects or endorses an option | A passing or negative mention does not establish it |
| Readiness score | A diagnostic of the site properties checked | It is not an answer-visibility measurement |

**Illustrative arithmetic:** if a brand appears in 3 of 12 eligible sampled answers, its mention occurrence rate for that sample is 25%. That does not mean 25% of people see the brand. The example is invented for explanation and is not an Omni Impact or customer result.

## Conditions needed for a publishable result

Record the prompt set; branded versus unbranded context; provider route and model/version; execution surface; dates; usable sample size; exclusions and failures; brand-detection rules; metric definition and denominator; competitor set; and any weighting or uncertainty method. Keep the underlying answers or a privacy-safe evidence extract available for fact-checking.

A confidence interval describes uncertainty under its assumptions. It does not remove prompt-selection bias, personalization differences, model changes or dependent observations. Repeated responses can be correlated. “No usable observations” is different from zero visibility.

## Comparing tools

Comparing two tools fairly means aligning their time windows, questions and engines. A comparison needs answer-level records, not just rows of citations or a screenshot of a score. Explain unmatched models, excluded rows and differences in prompt mix. A remaining gap is not automatically proof that either tool is inaccurate. No named competitor ranking or superiority claim is provided here.

## What the product does not establish

It cannot guarantee recommendations or reproduce every user's personalized consumer-app experience. It cannot infer customers, traffic, purchases or revenue from an answer count alone. A before/after observation does not prove that a particular edit caused a change. Training a foundation model, purchasing placement inside answers, and operating the third-party answer engines are outside the product description in this kit.