Operate
AI Citations
AI Citations is the rank tracker for answers. You tell Wrenda which questions matter and which brand is yours; it asks six AI models on a schedule, records what they said and who they cited, and scores your visibility over time.
Brands
Everything hangs off a brand (AI Citations → Add brand): the name AI models should be looking for, its aliases, and the domain it belongs to. Deleting a brand removes its prompts, responses, citations, competitors and battlecards, and cannot be undone. The Setup completeness checklist has two steps here: Citations brand (a brand exists) and Citations run (at least one run has completed).
Tracked prompts
Write prompts the way a buyer would ask an assistant, not the way you would type a search: “What is the best invoicing tool for a two-person design studio?” rather than “invoicing software”. A good starting set covers category questions, comparison questions (“X vs Y”), and problem questions your product solves.
| Plan | Tracked prompts | Cadence | Query quota / month |
|---|---|---|---|
| Starter | 50 | Weekly | 1,500 |
| Growth | 150 | Weekly | 4,000 |
| Pro | 300 | Weekly | 8,000 |
| Scale | 500 | Daily | 13,500 |
The six models
ChatGPT, Perplexity, Gemini, Google AI Overviews, Microsoft Copilot and Grok. Every enabled platform is queried for every prompt × persona combination on each run, with the full response captured so you can read exactly how each engine answered and which sources it cited. You can disable platforms under Settings → Citations; doing so reduces the queries a run consumes.
Cadence and quota
- Cadence. Weekly on every plan; daily on Scale. Settings → Citations lets you choose weekly, daily or manual within what your plan allows.
- Query quota. Each plan includes a monthly number of queries (a query is one prompt × one persona × one platform). A run reserves its queries up front; the Citations page shows the quota strip with used, remaining and the period.
- Configuration limits. Brands, prompts per brand, personas and competitors per brand are capped per plan; the settings page shows current use against each cap.
- Cooldown. Manual runs have a per-brand cooldown so a stray double-click doesn't spend the quota twice.
Personas
A persona re-asks every prompt from a point of view — CMO, SEO lead, founder, procurement — because models answer differently depending on who seems to be asking. Personas are managed on the brand's Personas tab and multiply queries (and quota) without changing the tracked-prompt count. Start with one or two; add more once you know which prompts matter.
Competitors and battlecards
Add competitor brands and their aliases under each tracked brand. Wrenda then separates competitor citations from yours in every response and computes share of voice: your mentions versus theirs in the same answers. From the Competitors tab you can generate a battlecard for a competitor — a comparison of share of voice and source gaps that shows where they are winning the answer and which sources are carrying them. Removing a competitor stops its comparisons and deletes its battlecard.
Opportunities
The Opportunities tab lists source gaps: third-party domains the models cite when answering your prompts but that never cite you. These are the pages worth getting listed on, quoted in or linked from, because they sit in the models' evidence pool. An empty list means either you already appear wherever your competitors do, or there isn't enough data yet — run more prompts first.
Scoring
Each response is scored on three axes and combined into a 0–100 visibility score per platform:
- Mention rate — how often your brand appears at all.
- Position — how prominent the mention is (opening paragraph versus footnote).
- Share of voice — your mentions versus competitors' in the same answers.
The brand overview charts the daily score per platform for the last 30 days.
Drift alerts
A drift alert fires when a platform's visibility moves by more than a threshold (default 15% week-over-week), in either direction, with a severity of info, warning or critical based on the size of the move. Alerts appear on the brand page and under Settings → Alerts, where you acknowledge them. Treat a critical drop the way you would a lost ranking: open the platform's latest responses and see who replaced you.
How runs work
Queueing
Scheduled runs start automatically at the cadence. A manual Run now validates the brand, claims the cooldown, reserves quota, and queues one job per prompt × persona × platform. The request returns immediately with the run marked Queued; a background drainer then works through the queue. If the quota is exhausted you get a clear error instead of a partial run; if the cooldown is active the error tells you how long to wait.
How long results take
The first batch of queries is fired straight away and the page polls every ten seconds while the run is queued or running, showing progress as responses land. Small runs typically complete within minutes; a large brand (hundreds of prompts × personas × six platforms) can take considerably longer because the queue is drained in bounded batches. A run ends as completed, partial (some queries failed — they are reported, not silently dropped) or failed.
Next
Pricing details for tracked prompts and quota are on the Pricing page.