Four metrics that answer different questions
- Mention
- An answer names your brand. It may do so without a link, without recommending you or in a negative context. Keep the surrounding answer.
- Citation
- An answer points to an identifiable source. Separate links to your website from links to third-party pages discussing your brand, and check whether the source supports the claim.
- Visit
- Your site records a session with an identifiable source where tracking allows. An observed citation does not establish that someone clicked it.
- Conversion
- A defined action is recorded, such as an enquiry, registration or purchase. A completed form is not automatically a qualified lead or a customer.
These can describe different parts of a journey. They do not, without additional evidence, establish a proven chain of cause and effect.
Read the scope before reading the trend
A percentage needs a denominator. Look for the following information before accepting the headline:
- the exact prompts and the reason they were selected;
- the market, language, tool, search mode and collection dates;
- the number of observations and repetitions for each prompt;
- the rules used to count mentions and citations;
- full answers or captures that make the examples auditable;
- protocol changes, collection errors and excluded observations.
Compare like with like. If new prompts are added, show the original panel separately from the expansion. Do not silently pool countries, languages and tools into a single visibility score.
A worked example, not a performance promise
Entirely fictional teaching example. These numbers are not Netcord customer data and are not performance targets.
Imagine 20 prompts, each observed five times for one tool and one language: 100 observations in each period. Both collections use the same protocol.
| Metric | Baseline | Next collection | Interpretation |
|---|---|---|---|
| Answers mentioning the brand | 30 of 100 | 42 of 100 | Up 12 percentage points within this panel. |
| Answers linking to the brand’s website | 8 of 100 | 12 of 100 | Up 4 points; this is not a click count. |
| Sales caused by the programme | Not established | Not established | These observations cannot establish that result. |
The supported conclusion is an observed increase within this sample. It does not establish improvement across every question in the market or isolate the agency’s contribution. Repeated answers may not be independent observations either: do not claim statistical precision that the collection method cannot justify.
Connect visibility to business outcomes carefully
Keep Reddit referrals, identifiable AI referrals and tracked conversions in separate views. Some journeys will remain incomplete or unattributed.
Google Analytics campaign parameters help identify campaigns on links you control. On their own, they do not demonstrate that the campaign caused a sale.
Google’s documentation on AI search features explains that their performance is included in Search Console’s Web search reporting. Do not relabel all organic growth as “GEO traffic”.
Ask which decision follows from the evidence: correcting a fact, improving a page, investigating a source or stopping an unhelpful activity. A useful report supports a next step, including a reasoned decision not to intervene.
A checklist for your next review
- I can inspect the prompts and observed answers.
- I know the tools, languages, dates and collection conditions.
- Rates include their denominators and exclusions.
- Mentions, direct citations and third-party citations are separate.
- Comparisons use a comparable scope.
- Reddit traffic, AI visibility and conversions remain distinct.
- Limitations and alternative explanations are explicit.
- Every recommended priority connects to an observation.
Keep this checklist for future reports. To understand the work behind the numbers, see our Reddit GEO method and the banking-sector case on our Reddit page. Its results belong to that programme, not a promised outcome for your market.