I do not care about an AI share-of-voice percentage until I can see the exact prompt, the competitor named instead, and the page the answer cited. These 12 copy-paste prompts turn groas’s 24/7 AI-answer tracking into that evidence inside your connected assistant; replace the square-bracketed placeholders and keep your tracked set fixed once you start.
Measure who wins without moving the goalposts
1. Compare monthly share against three fixed rivals
When I use it: I want the score and the losses behind it, not a percentage floating alone on a slide.
For the tracked prompts available through groas, compare the last 30 days with the prior 30 days for [BRAND] and [COMPETITOR 1], [COMPETITOR 2], [COMPETITOR 3].
Report:
- Total brand mentions used as the denominator.
- Mentions and share of voice for each named brand.
- The 5 prompts where our share dropped most.
- Exact prompt text and the competitors named for each drop.
State which brands are included in the denominator. Flag missing tracking periods rather than treating them as zero mentions.
Adjustment that matters: lock the competitor list and denominator. Changing the tracked competitors changes the shares, even when the underlying answers stay put. I want a comparison of performance, not a comparison of two different spreadsheets. Keep the same prompt set, engines, and reporting window for both periods.
2. Put visibility beside share of voice
When I use it: A report says I am “everywhere,” but competitors keep taking up the rest of the answer.
For [BRAND], using the same tracked prompts and date range as the previous report, show:
- Visibility: answers naming us / all answers.
- Share of voice: our mentions / all tracked brand mentions.
- Numerator, denominator, and percentage for both.
List 5 prompts where our visibility is high but our share is low. Include the exact prompt text and all brands named in the corresponding answers.
Adjustment that matters: keep both denominators visible. Forty answers naming you out of 100 means 40% visibility; 40 mentions out of 250 means 16% share. Neither number is broken. They answer different questions. A brand can appear regularly without dominating the conversation, and the next action depends on which problem the report actually shows.
3. Separate mentions, citations, and recommendations
When I use it: I need to know whether the answer merely names my brand or actually proposes it as a choice.
For tracked answers from the last 30 days, compare [BRAND] with [COMPETITOR 1], [COMPETITOR 2], and [COMPETITOR 3].
Show separate counts for:
- Mentioned: named in the answer text.
- Cited: the answer links to that brand's domain or page.
- Recommended: proposed as a choice to buy or shortlist.
List 5 exact prompts where we were mentioned but a competitor was recommended. Include the answer excerpt and any cited URLs. Mark ambiguous recommendations rather than forcing a classification.
Adjustment that matters: require all three columns. A mention, a citation, and a recommendation are different outcomes. Being linked while a competitor gets picked is not the same as winning the shortlist. I want to see which outcome is missing before asking for a rewrite.
4. Check the trend without chasing Tuesday
When I use it: A daily chart makes a jump look urgent, and I want to know whether the weekly pattern agrees.
For the same tracked prompt set, show weekly results for [BRAND] over the last 6 weeks.
Include:
- Share of voice.
- Our brand mentions and total tracked brand mentions.
- Visibility and total answer count.
- Engine coverage and any missing runs.
Highlight changes that persist across two full weeks. Show the exact prompts contributing most to those changes. Keep single-day jumps separate from the weekly trend, and do not label a change statistically significant without a justified method.
Adjustment that matters: compare like-for-like weeks, not isolated answers. I use a locked set of at least 30 unbranded prompts per market for this broader trend check, following the weekly measurement approach. Answer volume belongs beside the percentage. A thin week and a full week should not look equally convincing just because the chart draws both lines neatly.
Find the pages behind the citations
5. Rank cited domains and flag your presence
When I use it: I need the source list behind the score, with enough detail to decide what kind of work comes next.
For tracked [CATEGORY] prompts over the last 30 days, list the 15 most-cited domains by citation count.
For each domain, include:
- Citation count, first seen, and last seen within the period.
- Source type: owned / competitor / forum / review / editorial / media / blog.
- Whether any citing answer also cites [BRAND DOMAIN] or names [BRAND].
- One example prompt.
State which tracked engines are included. Mark unavailable fields rather than guessing.
Adjustment that matters: keep the source-type label on every row. A Reddit thread needs a different response from a competitor help document. The domain-and-source breakdown is the useful shape here: not five familiar logos, but a list I can sort into work. I look at review and forum rows first when investigating buyer prompts.
6. Drill down from domains to exact URLs
When I use it: A domain-level report points at a publisher but leaves me guessing which page actually matters.
For the top 5 cited domains in the previous output, list the 3 most-cited exact URLs per domain for [CATEGORY] prompts.
For each URL, show:
- Citation count.
- An example tracked prompt that cites it.
- What the page covers that [OUR URL] does not.
Compare page content only where you can access it. If either page is unavailable, flag the comparison as incomplete instead of inferring its contents from the title or domain.
Adjustment that matters: force URL-level output. Domain totals can hide several comparison pages doing the work. I want the page to inspect, not a publisher’s entire website as homework. This is the point of the citation gap audit: identify the specific source before deciding what deserves a rewrite.
7. Find cited sources where you never appear
When I use it: I want an inclusion list, not another inventory of places that already mention my brand.
From the last 30 days of tracked [CATEGORY] answers, find cited domains where none of the answers citing that domain also cite [BRAND DOMAIN] or name [BRAND].
Rank those domains by citation count. For each, include:
- Two exact example prompts.
- The cited URLs.
- A suggested action: pitch for inclusion / get listed / publish to replace.
- The reason for that suggested action.
Keep observed citation evidence separate from your recommendation.
Adjustment that matters: keep the absence filter strict. This is a source gap only when the answers citing that source do not also include the brand. The source-gap approach gives me places to investigate for inclusion, listings, or competing content. The suggested action is still a judgment call. A frequently cited domain is not automatically a sensible place to pitch.
8. Compare the same prompts across engines
When I use it: My brand appears in one engine and disappears in another, and I need to locate the difference before changing a page.
For the same 30 tracked prompts and the same week, split citation results by available engine: ChatGPT, Perplexity, Gemini, and Claude.
Show:
- The top 5 cited domains per engine.
- Where [BRAND DOMAIN] appears in one engine but is absent in the others.
- Exact prompt text and cited URLs for each gap.
- Any differences in available runs or web-search settings recorded in the data.
Mark an untracked engine as unavailable, not as zero visibility.
Adjustment that matters: compare the same week and actual coverage. Missing tracking is not missing visibility. When an engine gap appears, I check web-search settings and whether the cited page is fetchable before rewriting anything. The longer method is in finding competitor citation gaps in ChatGPT and Perplexity. Start with the conditions of the comparison, not a theory about a penalty.
Turn a tracked loss into a page change
9. Separate buyer questions from branded comparisons
When I use it: An overall score hides whether I lose category discovery, problem-led questions, or direct comparisons.
Group the last 30 days of tracked [CATEGORY] answers into:
- Category: best [CATEGORY] for [USE CASE].
- Problem.
- Comparison: [BRAND] vs [COMPETITOR 1].
- Informational.
For each group, show [BRAND]'s share of voice, the underlying mention counts, the top cited domain, and the exact prompt where we lose by the largest share margin to a competitor.
Keep branded comparison prompts separate from unbranded discovery prompts. Flag prompts that do not fit cleanly rather than silently changing the groups.
Adjustment that matters: lock the buyer questions before comparing weeks. For a smaller manual baseline, I use around 20 prompts across ChatGPT and Perplexity on the same day each week, within this 15-to-25-prompt approach. That is a smaller check than the broader trend set above. Do not mix the two and call the result one continuous trend.
10. Brief one page against one specific loss
When I use it: I have the losing prompt and cited URL, and I am ready to turn that evidence into a focused brief.
For this tracked prompt, [EXACT PROMPT], [COMPETITOR 1] was recommended and the answer cited [CITED URL].
Draft a content brief for [OUR URL OR NEW PAGE]. Include:
- The buyer question in plain language.
- Five subheads covering what the cited page answers and what it skips.
- Facts and proof we need to supply or verify.
- The relevant gaps in our current page.
- What not to rewrite.
Use the cited page and our page where accessible. Do not invent evidence, customer results, or product capabilities to fill gaps.
Adjustment that matters: use one prompt and one cited URL. Five losses in one request invite five thin outlines. I want a brief with a clear boundary: this buyer question, this competing source, this page to improve. “What not to rewrite” stays in the request because an evidence-led fix should not become an excuse to rebuild the whole site.
11. Stage the change with explicit permissions
When I use it: The brief is ready, and I want my groas-connected assistant to act without treating “draft this” as “publish whatever you like.”
Using the brief above, draft the replacement section for [OUR URL] in our voice.
Then stage it through groas at [SITE / PAGE PATH], under these rules:
[RULES: DRAFT ONLY; NO PUBLISH WITHOUT APPROVAL; STATE ANY OTHER LIMITS].
Show the proposed change and its reason before publication. Do not add unsupported claims. Log any change made through groas with its reason and retain one-step undo. If the requested action is unavailable, report that instead of claiming completion.
Adjustment that matters: put permission in the prompt itself. With groas connected to ChatGPT or Claude, the assistant can act on the site through groas; changes are logged with their reasons and can be undone in one step. I leave content autopilot off and approve each publication. groas Unlimited includes the tracking and site work for $19 a month, flat, with unlimited teammates, ad accounts, and websites. Permission still belongs in the request. Cheap mistakes remain mistakes.
The prompt I would copy first
12. List competitor wins with the cited page attached
When I use it: I want the shortest route from “our visibility looks weak” to a specific page worth investigating.
List tracked [CATEGORY] prompts from the last 30 days where [COMPETITOR 1] or [COMPETITOR 2] is recommended and [BRAND] is neither named nor cited.
For each loss, show:
- Exact prompt text and engine.
- Competitor recommended, with the relevant answer excerpt.
- Exact cited URL or URLs; mark answers with no citation.
- Source type: review / forum / editorial / competitor site.
- What the cited page states that [OUR URL] does not, where both pages are accessible.
Keep branded comparison prompts in a separate list. Rank the losses by how often they occur in the tracked period. Separate observed evidence from suggested page changes.
Adjustment that matters: keep branded prompts out of the discovery loss list. People typing the brand name are not the same audience as people asking which product to choose. I also read repeated losses over weeks, not one answer as a verdict. The limits of AI visibility reporting matter most when a tidy row starts looking more certain than the underlying answers.

This is my most useful prompt because it joins the loss to the evidence. Share scores start arguments about denominators. Citation tables start research projects. A row that says this prompt recommended that competitor through this URL, while the brand was absent, gives me something concrete to inspect. I would run this first, work through the top three rows, and let the other prompts keep score. Copy it today, lock the competitor list, and run it on the same day next week before changing a word on the site.

