Prompts are what zued sends to AI engines on your behalf. They are generated from your crawled page content and confirmed ICPs, then reviewed by you before dispatch. Their goal: to produce AI responses comparable to what your customers would see when asking about topics you cover — so you can measure how well your content matches what users are actually hearing.
How prompts are generated
Once your ICPs are confirmed, zued generates prompts per URL using your confirmed ICPs and the content of each page. Every prompt is ICP-driven — it reflects the situation, needs, and language style of the personas linked to that URL.
Since audit prompts run in neutral sessions — without chat history, memory, or personalisation — zued enriches them with ICP context to compensate. This makes AI engine responses comparable to what a real user with that background would receive.
Prompts are distributed across three context levels:
| Level | What it looks like |
|---|
| Minimal | A short question with one situational hint |
| Moderate | Several context details drawn from the ICP |
| Rich | Full ICP context with constraints, situation, and needs |
Intent goals
Each prompt is tagged with an intent goal that reflects the type of question a real user would ask:
| Goal | What it tests |
|---|
| Exploratory | Open research questions — “What should I consider when choosing X?” |
| Evaluative | Comparison and decision questions — “Which X is better for Y?” |
| Specific | Concrete questions seeking a direct answer — “What are the best X under Z budget?” |
Topic clusters
When your project contains multiple URLs covering the same topic, zued groups them into topic clusters. Prompts are generated per cluster using context from all related URLs, then each prompt is assigned to the single most relevant URL.
Topic names are normalised across languages and extraction runs. If you add URLs over time, new topics are matched against existing clusters — so “air fryers” and “Heissluftfritteusen” end up in the same cluster rather than creating duplicates.
Prompt-to-page fit
Every prompt is bound to one URL. When zued matches a prompt to your content, it picks the most relevant section of that page and shows you how confident it is in that match. The match confidence is shown as a label on each prompt and on the matched content card.
| Label | What it means | What to do |
|---|
| Strong match | The page clearly covers what the prompt is asking about. | Nothing. The audit is grounded. |
| Medium match | The topic is covered, but it isn’t the page’s main focus. | Optional: add a section that addresses the prompt directly, or accept that this URL covers the topic in passing. |
| Loose match | No section of this URL closely matches the prompt. | Move the prompt to a URL that fits better, or add the missing topic to this page. |
Prompt budget
Each plan includes a prompt budget of 7 prompts per URL. That works out to 70 prompts on the 10-URL plan, 175 on the 25-URL plan, and 350 on a 50-URL project. The budget is a shared pool across your project, not a fixed per-page quota: zued generates more prompts for pages that cover more ground and fewer for focused, single-topic pages, so the allowance flows to where it adds the most coverage. You review and approve every prompt before dispatch, and each approved prompt then runs on every engine in your plan, 10 times per weekly audit.
Brand exclusion
zued intentionally keeps your brand name out of generated prompts. The audit tests whether AI engines surface your content for a topic — not whether they already know your brand. This produces a neutral, repeatable baseline. Brand-specific prompts are only included when the page is explicitly about your brand (e.g. a comparison or “why us” page).
Reviewing and approving prompts
After generation, all prompts enter a review stage before dispatch. The review panel is organised by topic cluster, then by URL, so you can see at a glance which topics and pages are covered. You can:
- Edit prompt text, intent goal, or topic cluster
- Remove prompts that don’t fit your audit goals
- Add custom prompts to test specific queries you care about
- Import prompts in bulk from a CSV file
Once you approve, zued matches each prompt to your content and begins the AI dispatch. This review step ensures your audit covers exactly what matters to you.
Importing prompts from a CSV
If you already have prompts drafted in a spreadsheet, or you’re moving them over from another tool, you can import them in bulk instead of adding them one by one. The file needs a prompt_text column, and can optionally include url, topic_cluster, and primary_goal columns to assign each prompt to a page, cluster, and intent goal.
If you leave url, topic_cluster, or primary_goal blank, zued fills them in for you, matching each prompt to its most relevant page and tagging its intent, so a plain one-column list of questions still ends up fully organised. Anything you do provide is always kept as-is.
Imported prompts land in the review stage alongside generated ones, so you still confirm them before anything is dispatched. Before importing, zued shows you a preview: how many prompts will be added, and which rows were skipped (duplicates, empty or over-length text, or rows beyond your remaining budget) so nothing is dropped silently.
Custom prompts are dispatched and scored like generated ones. Keep generated prompts as the foundation — they’re systematically tied to your ICPs and content structure.
Managing prompts after approval
Approved prompts aren’t frozen. The All prompts view shows every prompt across all your URLs and clusters in one place, where you can filter by cluster, goal, or status and act on many at once:
- Add prompts, one at a time or by CSV import, directly to the live set. New prompts run at the next snapshot.
- Edit a prompt’s text, goal, or cluster at any time.
- Archive prompts you no longer want to track. Archived prompts stop running in future audits, while their past results stay visible in the snapshots where they ran. You can unarchive them later to bring them back.
- Merge or rename clusters by reassigning a group of prompts to a different cluster name. This is how you tidy up near-duplicate clusters (for example a singular and plural form of the same topic) or regroup prompts as your content evolves.
- Bulk actions apply any of the above to a multi-selected set, so you can reorganise a large project quickly.
When you change a prompt’s wording, the current snapshot keeps the result it already has, so your history stays intact, and the new wording runs at the next audit. The prompt is flagged so you know an update is pending.
This keeps your audit aligned with your content without breaking your week-over-week comparison baseline.
Why prompts include ICP context
Audit prompts run in neutral sessions — no chat history, no logged-in state, no personalisation. Without context, AI engines give generic responses that don’t match what your actual customers see.
ICP context compensates for this. By injecting situation details, constraints, and needs into the prompt, zued produces responses comparable to what a real user with that background would receive. This is what makes the coverage measurement meaningful: you’re comparing your content against the same answers your customers get.
Prompts and weekly snapshots
Prompts are generated once and then reused in every subsequent snapshot. They are never regenerated automatically. This is intentional: the same prompts running week after week is what makes trend tracking meaningful — you’re measuring the same thing each time, against fresh content and fresh AI responses.
Regenerating prompts resets your week-over-week comparison baseline. Only do this if your content or target audience has fundamentally changed.
