Track competitor visibility in AI answers by running a fixed prompt panel, declaring the competitor set before collection, and recording what each AI answer actually shows. In practical AI visibility tracking, the useful record is not a screenshot or one blended score. It is a row-level log that separates competitor presence, share of voice, share of citations, recommendation status, position or prominence, and visibility gaps.
The core question is simple: when a buyer asks an AI system for tools, alternatives, comparisons or recommendations, which competitors appear, which sources support them, and where is your brand missing or weakly framed? Do not answer that with one run or one score. Answer it by keeping prompts, answer surfaces, competitors, citations, and denominators stable enough for another reviewer to inspect.
The Short Answer: Track Competitors by Evidence, Not Screenshots
Competitor visibility tracking should start with evidence capture. For each prompt, save the answer text, answer engine, mode, date, market or language, visible citations, answer format, declared competitors, and observed competitors. Only then should you calculate share of voice, share of citations, or visibility gaps.
Use the report to make decisions, not to decorate a dashboard.
| Tracking layer | What to record | Decision it supports |
|---|---|---|
| Competitor presence | Which declared and observed competitors appear in the answer text | Whether competitors are occupying the answer surface |
| Recommendation status | Selected, favored, neutrally listed, caveated, dismissed or not applicable | Whether the competitor is merely named or actually preferred |
| Position or prominence | First in a ranked list, lower in a list, table row, supporting text or no clear rank | Whether the answer gives one competitor more attention |
| Share of voice | The competitor's share of counted visibility events in a fixed segment | Who is winning answer-level visibility against the declared set |
| Share of citations | The competitor's share of relevant visible citation events | Which brands or sources are supported by exposed evidence |
| Visibility gap | Competitors appear, get cited or get recommended while the tracked brand is absent or weak | Which prompt, source or positioning issue deserves inspection |
The important constraint is that each layer has a different denominator. Mention-based share of voice should not be mixed with citation share. Recommendation rate should not be calculated from educational prompts that never ask for a choice. Average position should not include paragraphs that do not have a real order.
Decision rule: if the finding cannot be traced back to a prompt, answer surface, competitor set, answer excerpt and denominator, treat it as an observation, not a benchmark.
Define Which Competitors Count
Before you measure competitor visibility, decide which competitors belong in the comparison. When competitor brand mentions and share of voice feed an AI brand tracker, the competitor set must be fixed before the answer capture starts. If you add brands after seeing the answers, share of voice and visibility gaps become unstable.
Use four competitor buckets. If the competitor list itself is still unresolved, decide how to pick competitors for AI brand tracking before calculating share of voice.
| Competitor bucket | Include when | Reporting caution |
|---|---|---|
| Direct competitors | They solve the same core problem for the same buyer in the tracked category | Excluding one can make the benchmark incomplete |
| Realistic alternatives | Buyers may choose them instead, even if the product category differs | Keep them separate from direct rivals |
| Category leaders | They shape broad discovery answers or define category language | Do not let them hide performance against closer competitors |
| Observed AI answer competitors | They appear unexpectedly in repeated in-scope answers, citations or recommendations | Log them separately before promoting them into the declared set |
This matters because AI-generated answers often do not follow a team's internal market map. A category discovery answer may include direct tools, broad SEO suites, review platforms, consulting options, marketplaces, or sources that are not competitors at all. A source domain can explain why a competitor appears without being a competitor itself.
Use this relevance check before adding a brand to the declared set:
| Check | Strong inclusion signal | Weak inclusion signal |
|---|---|---|
| Category fit | The brand solves the same problem or a realistic substitute problem | It only shares broad marketing language |
| Buyer overlap | The same user could compare both options for the prompt | Different buyer, workflow, budget or maturity |
| Prompt relevance | The brand belongs in the exact prompt group being tracked | It appeared once in a broad or unrelated answer |
| Market relevance | The brand is available and meaningful in the tracked market or language | It matters only in another region or segment |
| Decision impact | Its appearance changes shortlist, recommendation or positioning interpretation | It adds noise but no practical action |
Keep observed competitors in a separate field. If an unexpected brand appears once, save the evidence and monitor it. If it appears repeatedly in the same in-scope prompt group, review whether it belongs in the next declared set. Do not rewrite the active report while measuring it.
Build Prompt Groups for Competitive Intent
Competitor visibility appears in different ways depending on the prompt. A branded prompt can test whether an answer recognizes your brand, but it does not prove unprompted discovery. A recommendation prompt can reveal preference. A source-sensitive prompt can reveal citation gaps. Treat those as different measurement jobs. If the panel is not stable yet, first build prompt sets for AI rank tracking before treating the output as competitor data.
Start with prompt groups that map to buyer decisions.
| Prompt group | What it reveals | Example pattern |
|---|---|---|
| Category discovery | Which brands appear when no vendor is named | best [category] tools for [audience] |
| Alternatives | Whether competitors redirect demand away from a named brand | alternatives to [competitor] for [constraint] |
| Direct comparison | How the answer frames named options against each other | [brand] vs [competitor] for [use case] |
| Recommendation | Which brand is selected, favored or caveated for a scenario | which [category] tool should I choose for [need] |
| Problem-led | Whether the answer connects a user problem to a competitor set | how can I monitor [problem] in AI answers |
| Branded validation | Whether the answer describes a named brand accurately | what does [brand] do |
| Source-sensitive | Which sources, domains or page types appear around the answer | which sources compare [category] tools |
Do not mix prompt groups too early. A competitor may dominate alternatives prompts but barely appear in category discovery. Another may be cited often in informational answers but rarely recommended in buyer-intent prompts. If those results are blended into one AI visibility score, the report hides the practical issue.
The safest sequence is:
- Define the category, buyer, market or language, and answer surface.
- Assign each prompt to one prompt group.
- Lock exact prompt wording before collection.
- Run the same prompt panel across the chosen answer surfaces.
- Score each prompt group separately before summarizing.
If a prompt produces no brand set, no recommendation, no citations and no usable comparison, keep it as research context or remove it from competitor scoring. A broad educational answer is not automatically a visibility loss.
Capture Each AI Answer as a Row
A competitor visibility report is only as useful as its raw evidence. Each row should represent one prompt on one answer surface under one set of conditions. The row needs enough detail for another reviewer to understand why a competitor was counted.
| Field | What to record |
|---|---|
| Date captured | The date of the answer capture |
| Prompt | Exact prompt text, not a paraphrase |
| Prompt group | Category discovery, alternatives, comparison, recommendation, problem-led, branded validation or source-sensitive |
| Answer surface | ChatGPT, Gemini, Perplexity, Google AI surface or another declared environment |
| Mode and source visibility | Source-visible, search-enabled, model-only, localized or another stated condition |
| Market and language | Country, region or language when relevant |
| Declared competitor set | Competitors included before collection |
| Observed competitors | Brands surfaced unexpectedly by the answer |
| Answer format | Ranked list, unordered list, table, paragraph, source panel, hybrid or no brand set |
| Competitor status | Absent, named, shortlisted, selected, caveated, dismissed or not applicable |
| Position or prominence | First, lower in list, table row, supporting text only or no clear rank |
| Citation evidence | Visible URL, cited domain, source card, source panel or none visible |
| Source type | Own domain, third-party, review or directory, competitor-owned, generic or no visible source |
| Evidence excerpt | The sentence, row or bullet that supports the label |
| Action note | Monitor, rerun, inspect sources, update owned evidence, review comparison content or refine prompt |
Source-visible and model-only answers need different rules. If the answer surface exposes citations, citation share can be inspected. If it does not expose sources, the answer can still support presence, recommendation, competitor, and framing analysis, but it should not be counted as a citation failure.
Answer format also matters. A numbered list can support position if the order is meaningful. A table may show inclusion without a clear winner. A paragraph can mention several brands without ranking them. Do not force every answer into average position just because the dashboard has a position field.
Red flag: a report counts a competitor's cited URL as a competitor mention even though the answer text never names the competitor. That is source exposure, not answer-level presence.
Measure Share of Voice Without Hiding the Components
AI share of voice is useful when the competitor set, prompt set, and counted event are fixed. It should answer a specific competitive question: which brands take the largest share of relevant AI answer visibility in this segment? It should also sit beside other AI visibility metrics, not replace them.
The simplest safe formula is:
AI share of voice = a brand's counted visibility events divided by all counted brand and competitor visibility events in the same prompt set, answer surface and reporting window.
The counted event must be defined before reporting. Different teams and tools may count mentions, qualified mentions, recommendations, placement-weighted visibility, or another stated event. That is fine if the report exposes the rule. It is misleading if the rule is hidden.
Use separate share views instead of one opaque score.
| Share view | Counted event | Best use | Main risk |
|---|---|---|---|
| Mention share of voice | Brand or competitor named in answer text | Basic competitive presence | Counts weak mentions unless labels are strict |
| Qualified visibility share | Shortlisted, prominent or meaningfully evaluated appearances | Cleaner view of answer-level consideration | Requires consistent reviewer rules |
| Recommendation share | Selected or favored recommendations | Buyer-intent prompts where a choice is requested | Should not apply to neutral educational prompts |
| Position-aware share | Prominence in ranked or clearly ordered answers | Lists where order matters | Creates false precision if paragraphs and tables are included |
| Segment share | Share by prompt group, engine, market or language | Finding where the competitor issue actually appears | Needs stable segments before summarizing |
Read share of voice by segment before averaging. A competitor may have strong share in alternatives prompts because users ask about that competitor directly. That does not mean the same competitor dominates unbranded category discovery. A brand may appear in ChatGPT-style answers but not in another answer surface. A single total can hide the reason.
Use share of voice when these conditions are true:
- The category and prompt group are in scope.
- The declared competitor set is fixed for the reporting window.
- The answer surface and mode are labeled.
- The counted visibility event is defined.
- The denominator is visible.
- Raw answer evidence is available for review.
If those conditions are not true, report competitor observations instead of share of voice. Observations can still be useful, but they should not be presented as trend-ready metrics.
Measure Share of Citations Separately
Share of citations answers a different question from share of voice. Share of voice measures answer-level visibility. Share of citations measures visible source evidence. The distinction between AI mentions and AI citations matters here: a competitor can be mentioned without being cited, and a competitor can be cited as a source without being recommended. A third-party source can cite or discuss competitors without being a competitor itself.
Keep citation analysis inside source-visible answers.
| Citation case | What happened | How to interpret it |
|---|---|---|
| Mentioned and cited | The answer names the competitor and exposes a related source | Inspect the source, claim and recommendation context |
| Mentioned but not cited | The competitor appears in answer text without visible source evidence | Treat it as presence, not citation strength |
| Cited but not mentioned | A competitor-owned or competitor-related URL appears, but the answer text does not name the competitor | Treat it as source exposure, not competitor presence |
| Competitor cited, tracked brand absent | A competitor or competitor source gets evidence while the tracked brand is missing | Possible citation gap if the prompt is in scope and the pattern repeats |
| No visible citations | The answer provides no source evidence | Do not draw citation-share conclusions from that run |
Classify cited sources before deciding what to inspect.
| Source type | What it may explain | Practical next step |
|---|---|---|
| Own-domain source | Whether controlled pages support the brand clearly | Check category, use case, comparison and claim clarity |
| Third-party source | Whether external coverage includes or omits competitors | Inspect lists, reviews, directories and editorial summaries |
| Review or directory source | Whether sentiment, fit and buyer expectations are shaped externally | Check profile completeness, category labels and outdated facts |
| Competitor-owned source | Whether competitors are framing the category or comparison | Review the specific claim before deciding whether to publish clearer evidence |
| Generic source | Whether the answer is using background material rather than brand evidence | Do not over-prioritize unless it repeats across important prompts |
The share of citations denominator should be based on relevant visible citation events, not all prompt runs. If one answer surface does not expose citations, keep it out of citation-share reporting. It may still belong in presence or recommendation reporting.
Decision rule: use share of citations to decide which source layer to inspect. Do not use it to claim that a competitor was preferred unless the answer text also shows recommendation evidence.
Find Visibility Gaps That Deserve Action
A visibility gap is not just "a competitor appeared." It is a pattern where competitors appear, get recommended, or receive visible source support in an in-scope prompt while the tracked brand is absent, weakly framed, unsupported, or incorrectly described. When those patterns repeat by topic cluster, they can become AI brand tracking topic gaps rather than isolated competitor notes.
Use this step-by-step process before acting.
- Confirm prompt fit. Is the prompt inside the category, buyer segment, market and intent you care about?
- Check the competitor set. Did a declared competitor appear, or is this an observed competitor that needs review?
- Identify the signal. Is the gap about presence, recommendation, position, citation, framing or accuracy?
- Inspect the answer format. Does the answer support a rank, a recommendation label, a citation label or only a plain mention?
- Check source visibility. Are citations visible, and do they support the competitor claim?
- Look for repetition. Does the same pattern appear across the same prompt group, surface or market?
- Choose the action. Monitor, rerun, inspect sources, update owned evidence, review comparison content or refine the prompt panel.
Different gaps point to different work.
| Gap pattern | Likely interpretation | Better next action |
|---|---|---|
| Competitors appear in category discovery, tracked brand absent | Weak unprompted discovery or category association | Inspect category evidence and relevant third-party lists |
| Competitor selected while tracked brand is listed neutrally | Visibility exists, but recommendation strength is weak | Review comparison evidence, use-case fit and caveats |
| Competitor cited repeatedly, tracked brand uncited | Source evidence may favor the competitor | Inspect cited source types before changing content |
| Competitor-owned pages appear in source panels | Competitor framing may influence the answer surface | Map the claim, then decide whether clearer owned comparison evidence is needed |
| New competitor appears repeatedly across prompt groups | The answer environment may reveal a real competitive pattern | Keep it observed first, then review for the next declared set |
| No brands appear in the answer | The prompt may be educational or too broad | Refine or remove the prompt from competitive scoring |
The action should match the signal. If the gap is citation-based, inspect sources. If the gap is recommendation-based, inspect positioning and comparison evidence. If the gap appears only once, monitor or rerun before escalating. If the same declared competitor appears across category discovery, alternatives, and recommendation prompts while the tracked brand is absent, that is a stronger visibility gap than a one-off omission.
Red Flags Before Reporting Competitor Visibility
Competitor visibility reports often look precise while answering the wrong question. Check these red flags before sharing the result.
- The competitor set changes mid-report: share of voice, recommendation share and citation share no longer have stable denominators.
- Declared and observed competitors are blended: a surprise answer changes the benchmark instead of informing the next cycle.
- Branded prompts are treated as discovery: recognition after the prompt names the brand is not the same as unprompted visibility.
- Mentions are counted as recommendations: a named competitor may be listed neutrally, caveated or dismissed.
- Citations are counted as mentions: a cited URL is not answer-level competitor presence unless the answer text names or evaluates the competitor.
- Source-visible and model-only answers are averaged: citation conclusions become unreliable when surfaces expose different evidence.
- Every answer is forced into rank: paragraphs, tables and neutral lists do not always support numeric position.
- No raw answer archive exists: another reviewer cannot inspect the prompt, answer excerpt, citation or label.
- One run becomes a trend: normal answer variation can look like competitive movement without repeated evidence.
- A composite score hides the cause: the report cannot say whether the movement came from presence, recommendation, citations, prompt mix or competitor-set changes.
Do not report competitor AI visibility as a final competitive conclusion when the category boundary is unclear, the prompt panel is exploratory, the answer surface is changing, or the team cannot act on the finding. In those cases, the right output is a measurement note: what appeared, why it might matter, and what must be stabilized before the next run.
Practical Takeaway
Tracking competitor visibility in AI answers is a controlled benchmark, not a search for interesting screenshots. Define the competitors before collection, run stable prompt groups, capture each answer as evidence, and keep the signals separate.
Use share of voice to understand answer-level competitive presence. Use share of citations to inspect visible source evidence. Use recommendation status and prominence to see whether competitors are merely named or actually favored. Then use visibility gaps to choose the next action: monitor, rerun, inspect sources, improve owned evidence, review comparison content, or refine the prompt panel.
The strongest competitor visibility report is not the one with the biggest score. It is the one that can explain which competitor appeared, in which AI-generated answer, for which prompt, with which citation evidence, and what decision should happen next.
