Validation Tracking — Prominara Docs

Validation tracking monitors when AI platforms mention your brand in their responses to user queries.

Building Your Prompt Set

/validation/new is a set-building workbench, not a blank form. It opens on a list of questions drafted from your own site, with a balanced starting set already selected, and a sidebar that shows the set as you shape it. Three ways in, one set out:

Suggested for you — the drafted questions. Keep the ones that sound like your customers, drop the ones that don't, edit any of them in place.
Write your own — type a question the way a customer would ask it.
Paste or upload — bring a list from a doc or spreadsheet.

Then pick platforms and cadence, review, and create:

Navigate to Validation in the sidebar and click "New Validation"
Shape the set — the sidebar tracks slots used and which buying moments are covered
Select which platforms to check (ChatGPT, Perplexity, Google AI (Gemini), Google AI Overviews, and Google AI Mode)
Choose check frequency, then review and create

Onboarding uses the same workbench, so the set you start with on day one and the set you extend six months later are assembled by the same rules.

The Recommended Set

The pre-selected set is sized to your plan's prompt slots and balanced on two axes at once: the three kinds of question (Discovery, Your brand, Comparison) and the moments of the buying journey — from someone exploring a problem, through looking for options and comparing choices, to ready to buy. Discovery gets the largest share, because those are the only questions that count toward your headline visibility rate.

You are never locked into it. "Use recommended set" restores the balanced selection at any point; everything you wrote or pasted yourself is left alone.

When Prominara doesn't know any competitors for your brand yet, the comparison slots fold back into Discovery rather than inventing rival names to put in front of an AI.

Batch Creation

The Batch tab lets you create up to 100 prompts in a single request — ideal when migrating a tracking list or kicking off a new engagement. Two input modes are supported:

Paste - One prompt per line. Plain text, comma-separated, or enrichedtext | category | intent format (category/intent optional).
CSV / TXT upload - Columns are auto-detected fromprompt, category, intent headers — column order doesn't matter. A downloadable template is provided inline.

The form previews every row, flags problems line-by-line ("too short", "not a valid intent", duplicates), and dedupes against prompts already tracked for this site before submitting. Prompt uniqueness is scoped per site, so the same question can live under two different brands in the same organization without collisions.

Batch creation respects your plan's prompt and monthly AI-check limits. If the batch would exceed either limit, the form blocks submission with a clear message and a link to upgrade.

Validation Components

Prompt

The question or query to ask AI platforms. Write prompts the way real users would ask.

Good: "What are the best tools for tracking AI citations?"
Bad: "Tell me about Prominara company"

Question Type & Buying Moment

Every prompt is filed as one of three types — Discovery (never names your brand), Your brand (names you), or Comparison (names a rival) — and Discovery prompts also carry the buying moment they sample, from exploring a problem through to ready to buy. Both are set for you when the prompt is created and shown as badges in the prompt list, which you can filter by type.

Only Discovery prompts count toward your headline visibility and citation rates. A question that already contains your name will hand it back to you, so including those answers would inflate the rate. Brand prompts feed accuracy and sentiment; comparison prompts feed competitive positioning. Each panel is reported on its own terms — never averaged into a single number. See Citation Tracking.

The type is frozen at creation and recomputed only when you edit the question text itself, so a rate can never quietly change meaning between two weeks of history.

Category & Intent (Optional)

Tag prompts with a category (e.g. comparison, how-to, product discovery) and intent (informational, commercial, transactional) to segment results in reports. Both are optional — untagged prompts still work and the empty state collapses cleanly.

Platforms

Select which AI platforms to query:

ChatGPT - OpenAI with web search
Perplexity - Perplexity Agent API
Google AI (Gemini) - Gemini with live Google Search grounding
Google AI Overviews - Real SERP fetch (DataForSEO), tracks the AI summary that appears at the top of Google Search
Google AI Mode - Real SERP fetch (DataForSEO), tracks Google's conversational AI answer surface (opt-in, English-language markets)

Brand Assignment

Validations are scoped to the brand selected in the global brand picker (sidebar orCmd/Ctrl + B). Switching brands swaps the active site without leaving the page, and the URL pins to /b/[siteId]/validation so links you share stay scoped to the right brand.

Set Health

Once prompts are tracked, the Validation page carries a health strip and a How it's built tab that read the library the same way the workbench does: slots used against your plan, how many buying moments are covered, and how the three question types stack up against their targets. When something is missing you get one action sentence, not a chart to interpret.

Inside each type, suggestions are ordered by an opportunity score — a composite of the question type, overlap with real Google Search Console queries your site already gets impressions for, and how well the question fits the topics, audiences, and competitors detected during scans.

Suggestions are ranked by opportunity score across the validation wizard, onboarding, and the API — the same order everywhere.

Measured Prompts

Prompts you mark for measurement are sampled instead of checked once: each is run repeatedly per platform, and against frozen alternate phrasings of the same question as well as the original. The phrasings are generated once and never rewritten, so week-over-week comparisons stay valid.

Sampled prompts carry an answer stability badge per platform — stable, variable, or volatile — derived from whether identical runs agree with each other and how far apart the different phrasings land. A prompt cited on every check but flipping on reword is shown as volatile rather than counted as a win.

Unmarked prompts stay on routine single-run monitoring. The extra sampling runs are not billed against your monthly AI checks; only the primary check counts. Stability badges only appear where a prompt was actually sampled. The full protocol is published at Methodology.

Interpreting Results

For each validation run, you'll see:

Citation status - Green check (cited), red X (not cited)
Mentioned vs. cited - Whether your brand was named in the answer text, linked as a source underneath it, or both — tracked separately, never merged
Retrieval - Whether the engine searched at all on this run, which separates "never looked" from "looked and chose a competitor"
Sentiment - Positive, neutral, or negative mention
Competitors mentioned - Other brands in the response
Full response - The complete AI answer
Run timestamp - When the validation was executed

History Tab

The History tab groups every platform check that belonged to the same trigger into a single run, so a 3-platform scan always shows as one row (3/3) rather than being split if a single platform takes longer to respond. Citation-rate trending compares the most recent runs against the previous window.

Citation KPIs & Trend

The Validation page header surfaces Coverage (share of tracked prompts that returned at least one citation) and Citation rate (cited responses divided by total platform runs in the window) as headline KPIs. Both count only Discovery prompts — questions that never mention your name — because those are the only ones where a citation is evidence you were found. Each card also shows the raw counts behind the ratio — how many distinct validations were cited and how many were checked inside the reporting window — so you can tell the difference between a small sample and a stable trend at a glance. Hover the info icon on either card for the exact formula. The citation-trend sparkline shows the rolling delta rounded to one decimal place so small movements stay legible.

Brand and comparison prompts are still checked, stored, and shown — they just live in their own panels. Filter the prompt list by type to read each one on its own terms.

Keep all three types in your set, and spread the Discovery ones across the buying journey. A set that only asks shortlist questions ("best tool for...") will miss the buyer who is still describing their problem.