Understanding the Scores

ReachLLM runs your tracked prompts against each enabled AI platform on a schedule (daily or weekly depending on your settings) and analyzes every response. This page explains each metric that comes out of those runs: what it measures, how it is computed, and what moves it.

Choose the comparison group

Brand Visibility uses the competitors marked Direct in Knowledge Base by default. The same comparison group is used for Share of Voice, Average Rank, competitor trend lines, the Comparison Matrix, prompt gaps, and generated reports.

Choose All tracked to include every company found in the analyzed answers, including companies classified as Other. If a brand has no direct competitors set, ReachLLM clearly falls back to All tracked instead of showing an unexplained own-only score. The selection applies to competitor comparisons, including Share of Voice, rank, competitor trends, the Comparison Matrix, Gap Analysis, detailed-response competitor summaries, and reports. Visibility Score plus your own mention and citation rates do not change when you switch comparison groups. General source analytics is also independent because it describes every source cited in the selected AI answers.

Free reports add one evidence safeguard to this rule. Selected direct competitors remain primary, but if fewer than three have measurable appearances, the published report supplements them with the highest-evidence companies already found in the same analyzed answers. These alternatives are never invented. If the answers contain fewer than three evidenced competitors in total, the report labels the cohort as Limited comparison evidence and does not show a 100% result as representative market share.

Visibility Score

What it measures: how often a platform's answers mention your brand at all.

How it is computed: for each platform, the Visibility Score is the percentage of your tracked prompts where that platform's answer mentions your brand. The overall Visibility Score is the average across platforms that ran at least one prompt.

Worked example: you track 20 prompts. ChatGPT mentions your brand in 12 of the 20 answers, so your ChatGPT Visibility Score is 12 / 20 = 60%. If Gemini scores 50%, Perplexity 40%, and Google AI Overviews 70%, your overall Visibility Score is the average of the four: (60 + 50 + 40 + 70) / 4 = 55%.

How to improve it: make sure your tracked prompts reflect the questions your buyers actually ask (see Working with Prompts), fix the issues your GEO audit surfaces so AI platforms can read and trust your pages, and publish content that directly answers the prompts where you are absent. The Responses tab shows the raw answers, so you can see exactly which prompts you are missing from.

Topic-wise Visibility

Topic-wise Visibility breaks the selected result set into the topics assigned to its prompts. For each topic and brand, ReachLLM divides the number of successful tested AI answers that mention that brand by the total successful tested answers for the topic. A brand counts at most once in an answer. Failed checks and explicit no-trigger results are not presented as tested answers.

For example, a topic contains three prompts tested across four AI platforms, producing 12 successful answers. If eight answers mention your brand, its topic-wise visibility is 8 / 12 = 67%. A prompt assigned to two topics contributes its answers to both topic calculations.

This topic percentage is a cumulative answer-level breakdown, while the headline Visibility Score first calculates a rate for each platform and then averages the active platforms. The values are usually close when every platform returns the same number of answers, but the calculation is intentionally labeled separately. Prompt and provider filters change the tested-answer set. The comparison-group control changes the competitor rows, not your brand's own topic percentages.

What it measures: your slice of the conversation compared with competitors.

How it is computed: across all analyzed answers, ReachLLM counts every brand appearance in the selected comparison group, then reports the percentage that are yours. It is presence-based: a brand counts once per answer, no matter how many times it is repeated within that answer. Direct competitors are selected by default. All companies is available when you want the broader market view.

The Share of Voice detail ranks your brand and competitors that earned a measurable share. Companies at 0% are omitted from the ranked table, while the comparison-group description still explains which cohort the calculation used. The table shows the number of answers that mention each brand and labels changes in percentage points.

Worked example: across a run, the analyzed answers contain 40 brand appearances in total. 10 of those are your brand and 30 belong to competitors. Your Share of Voice is 10 / 40 = 25%. If one answer repeats your brand name five times, that still counts as a single appearance for you in that answer.

How to improve it: Share of Voice is relative, so it rises when you appear in more answers or when competitors appear in fewer of the answers you both compete for. Target the prompts where competitors show up and you do not, and keep your competitor list accurate: competitors are auto-discovered during brand analysis and fully editable, including direct, indirect, SERP, and Other Company classifications plus aliases for alternate spellings. Wrong mentions can be removed and metrics recomputed, which keeps this score honest.

Average Rank (Visibility Rank)

What it measures: when your brand does appear in an answer, how early it is mentioned relative to the selected comparison group. Lower is better.

How it is computed: within each answer, selected brands are ranked by their first mention. Average Rank is the mean of your positions only in answers that mention you. An absent competitor receives no rank. Answers where your brand does not appear are not part of this calculation.

Worked example: your brand appears in three answers. In the first it is the first brand mentioned (position 1), in the second it is second (position 2), and in the third it is third (position 3). Your Average Rank is (1 + 2 + 3) / 3 = 2.0.

How to improve it: being mentioned is step one; being mentioned first is the goal. Strengthen the signals that make AI platforms treat you as the primary answer: a complete brand knowledge base, authoritative content on the topics behind your prompts, and off-page authority (see the off-page checks in the GEO Audit). The trends charts show rank history per platform and per brand versus competitors, so you can confirm movement over time.

Sentiment

What it measures: the tone of what AI platforms say when they mention you.

How it is computed: ReachLLM checks every saved answer from each platform. Answers that identify your brand receive a complete-answer sentiment assessment supported by attributable quotations. Answers without the brand are checked but excluded from sentiment scoring. Comparisons retain which business each statement belongs to. Shared or comparative context can support an answer assessment while staying separate from company-specific claims. Each assessed brand answer contributes 100 for positive framing, 50 for mixed or neutral framing, or 0 for negative framing. A positive sentence on its own does not establish that the whole answer is positive. Mixed answers retain qualifications such as pricing concerns or suitability limits. The platform score is the average of its assessed answers, rounded to a whole number. The overall score weights platform scores by their assessed answer counts and is also rounded.

Worked example: ChatGPT has four assessed answers: two positive and two mixed, for a score of 75. Gemini has two assessed answers: one positive and one mixed, also scoring 75. The overall score is (4 x 75 + 2 x 75) / 6 = 75. Missing or rejected assessments contribute no score, rather than a zero or an assumed neutral result.

This index summarizes the tone of the assessed brand answers. It is not a customer review score or a measure of medical, product, or service quality. An explicit overall rating stated by a provider, such as 7/10, is preserved separately with its exact supporting passage. App-store ratings, review averages, and conditional ratings are not substituted for the sentiment index.

Per-platform coverage distinguishes saved answers checked from brand answers assessed. For example, 15 saved answers can contain five brand answers: complete coverage means all 15 were checked and all five brand answers assessed, not that the other ten received neutral scores. Missing text, uncertain business identity and failed assessments are identified separately. Recoverable unfinished work resumes automatically, revisiting the specific evidence behind a rejected assessment and retaining successful assessments; persistent gaps remain partial or unavailable rather than receiving invented scores.

Scheduled coverage is checked against the platforms and prompts recorded for that run. A platform or prompt that saved no answer appears as missing, including when every request for a platform failed. Missing scheduled answers do not increase the number of saved answers checked. If an older run no longer has its original platform or prompt roster, ReachLLM can finish reviewing its recorded answers but labels the original scheduled coverage as unavailable. It does not assume that today's platform settings were used historically.

Topics come from the business-specific evidence, so their number varies by brand and run. Compact citation labels keep summaries readable. Evidence views retain the original quotations and sources. A score of 100 describes the assessed answers, not every customer's experience. Reports using the earlier selected-quote or capped-sample methods are not directly comparable with a full saved-answer assessment. Saved reports retain the scores and coverage available when they were created.

Sentiment uses the completed full analysis selected for the current prompt inventory. In-progress or smaller exploratory runs do not replace it. Legacy date formats are compared as actual dates, so an older stored result cannot hide a newer matching analysis.

When a provider retry changes an answer, its previous sentiment is withdrawn while ReachLLM reviews the updated saved evidence. Existing successful assessments for unchanged answers are reused. An answer that fails its first evidence review can receive one stronger second review. Both reviews use the same source and attribution checks; missing evidence is never invented. A failed or unresolved assessment remains visible as incomplete rather than receiving an invented score.

Shared category membership is neutral context. A comparison or a supported technology relationship may help assess the complete answer without turning another company's features into your brand's strengths. If an evidence-handling correction makes a previously rejected passage valid, only that affected answer can receive one further review. Unchanged failures do not restart.

New reports use the latest saved sentiment assessment. An unavailable score stays unavailable throughout the report, including provider comparisons; an older visibility metric cannot replace it. A measured score of zero remains zero. Original evidence and qualification text remain available in the report. The original analysis date and its sentiment review date are shown separately from the report creation date. Unknown source dates remain unavailable.

Head-to-Head: this card always compares your brand with one competitor marked Direct in Knowledge Base. It shows two independent mention rates because the same answer can mention both brands. The difference is labeled in percentage points. Competitor sentiment is not shown because it is not evaluated with the same method as your brand's sentiment.

How to improve it: read the drivers in the Sentiment tab to see which themes pull negative or neutral. Then address the source material: customer reviews, comparison content, and the pages AI platforms draw from. Content that clearly states your strengths, backed by your knowledge base, gives platforms more positive material to work with.

Citation rate

What it measures: how often AI platforms cite your own domain as a source. This is different from being mentioned: an answer can name your brand without linking to your site, and it can cite your site without naming you prominently.

How it is computed: for each platform, the citation rate is the percentage of triggered prompt runs where the answer cited your domain as a source. A failed triggered run counts as not cited. The overall citation rate is the average across platforms that ran at least one prompt.

Worked example: across 100 prompt runs on a platform, your domain appears as a cited source in 8 of them. Your citation rate on that platform is 8 / 100 = 8%.

How to improve it: citations go to pages that are easy for AI platforms to retrieve and quote. Use the GEO Audit to fix on-page issues like structured data, snippetability, and llms.txt, and check the Sources tab to see which domains AI platforms cite for your prompts today. PR Outreach helps you earn placements on the publications that already get cited, and publishing content that directly answers your tracked prompts gives platforms a reason to cite you instead.

Sources, opportunities, and Gap Analysis

Sources & Opportunities opens with source analytics. Use it to understand every cited domain and URL, source categories, and how citations change over time. This source library is not a competitor comparison.

Choose the optional Gap Analysis view when you want questions where a competitor in the selected comparison group appears and your brand does not in the same AI answer. Brand not mentioned and Website not cited are identified separately. Each row shows compact competitor logos, AI platforms, and the number of affected answers. Focus or hover a logo to see the company name. You can review the exact answers, open related queries, or choose Plan a fix to ask the Agent for the content or authority work that can close the gap.

When Gap Analysis is open, the source analytics below it narrows to sources cited in those exact gap answers. Return to Source analytics to see every publication, forum, directory, and domain used across the analyzed answers.

Where to see these metrics

The results views include five tabs: Overview, Sources and Opportunities, Sentiment, Query Fanout, and Responses. Trends over time charts show score, share of voice, and rank history per platform and per brand versus the selected comparison group. In the platform trend view, Prompt coverage means the share of tracked prompts where at least one AI model mentions or cites you. It is different from Visibility Score, which averages each selected platform's mention rate. Information icons beside the headline metrics show the calculation in the app and can be opened with a pointer or keyboard.

Generated reports preserve the comparison group, tracked prompt count, AI model count, metric definitions, and reporting date used when the report was created. Changes are labeled in percentage points. This context stays frozen with the report so a later competitor change does not silently rewrite an older result.

Next steps

AI Visibility Tracking: how tracking runs work end to end.
Working with Prompts: tune the prompt set these scores are built on.
GEO Audit: fix the on-page and off-page issues holding your scores back.