CitadexDocs

Prompt Library: collect the questions customers really ask AI

Find candidate questions from real searches, see whether your prompts cover enough, and add and import the prompts you track: how the Start from real searches card, Coverage and candidates, and the Quick Test, Import and Add Prompt buttons work, and how prompts use credits.

Product Manuals19 min read

What this page is for

Citadex reports only on the prompts you track, so your prompt set defines what you can see. The Prompt Library finds the questions worth tracking from each market's real searches, shows whether your prompts cover each kind of question, buying stage, topic and customer role, and is where you add and import prompts and manage markets. The prompts you track are listed and managed in Prompt Tracking.

When to use it

  • After creating a project, to grow the generated prompts into a representative set.
  • When a group of customer questions (a city, a competitor comparison) isn’t covered.
  • After adding a market that needs prompts in its language.

The page

Top bar: on the right, Quick Test (opens the test dialog), Import (opens the import dialog), Markets & engines (folded into ⋯ on a narrow screen) and the dark Add Prompt (opens the same dialog as Add Prompt on Prompt Tracking, with Market set to the market chosen on this page; you can paste many prompts at once, and question types are assigned automatically; see Add Prompt (top bar) below). Below it are the Start from real searches. card and the Coverage and candidates table, with no tabs; the prompts you track are listed in one place, Prompt Tracking. The former My Prompts tab was retired: old links (?tab=myPrompts) open Prompt Tracking, filtered to the market when the link names one (&market=); ?markets=1 still opens the add-market dialog. The former coverage tab is now part of this page too; old links (?tab=coverage) open the Prompt Library.

Find candidates and check coverage

Find the questions customers ask in real searches, see whether each kind of question is covered well enough, and tick candidates to track them.

  1. Top row: the current market by its country alone, e.g. United States, with the language added only when the project monitors that country in more than one language, e.g. Canada · English (the market every source searches in; there is always one market, switched from its menu; hovering shows the full Market: United States · English), with 文A on the right.
  2. Start from real searches. card: under the title, "Find new candidate questions from your site’s searches, search suggestions and questions people ask in public."; on the right, the dark Find candidate questions. It takes real searches in this market (your site's Search Console, Google autocomplete, "People also ask"), rewrites them as questions people ask AI assistants, and sorts them by kind of question into Coverage and candidates below. Only a kind of question that no real search fits gets a few written by AI, marked AI-generated. Once Search Console is connected, your own site's searches are used automatically.
  • Or pick a source: six buttons, each with its own coloured icon. Each opens a small dialog (AI engine searches needs none), and what it finds goes into the candidates below, never straight into tracking:
  • Import from Search Console: pick queries from your site's Search Console, then Add to candidates n. The list shows the searches in this market's country; the demand heat of the picked queries is graded on the Search Console impressions Citadex reads for this market's country.
  • Expand suggestions: up to 5 seed words, for this market's Google autocomplete and "People also ask".
  • Reddit / forums: a topic, for questions people asked in their own words on Reddit and Quora.
  • Keywords → prompts: your own keywords rewritten as questions people ask AI.
  • From a competitor URL: a page URL; AI infers the questions the page is written to answer. These have no search evidence, and their candidate pill says Inferred by AI.
  • AI engine searches: reads this project's scans of the last 30 days and lists the web searches AI engines ran before answering your tracked prompts that recur and are not tracked or listed as candidates yet, as the engines wrote them, with the engines that ran each and how many answers it appears in. They are the engines' own searches, not questions people asked, and they do not count as demand evidence, so they are not ticked. Some AI engines expose no searches; the notice lists them. When new ones appear, the AI Assistant lists them in the chat at most once a week, with one card; confirming it adds them as candidates under “Coverage and candidates”, not tracked and not ticked.
  • Or expand a seed: type a question, or Choose a tracked question (commercial-intent ones are listed first), then click Expand: real searches around it first, then AI fills in, about 10 in all; anything AI adds is marked.
  • Finding candidates, from any source, uses no credits; you can do it a limited number of times a day. AI engine searches reads existing scans only and does not count toward that number.
  1. While it looks, and after: after the dark button, a source or Expand, a line under the card says "Looking through this market’s real searches, usually a few seconds…". Then a green notice says how many candidate questions were found (how many from real searches, how many written by AI) and that they were added under their kind in Coverage and candidates, marked New; View candidates jumps to the table, and the page also scrolls down to it. When nothing new turns up, it says so and suggests another source or expanding a question.
  2. Coverage and candidates: laid out like the Prompt Tracking list.
  • Toolbar: one row, as on Prompt Tracking. On the left, an icon and Coverage and candidates · Candidates N, with an ⓘ that reads "Under each kind, the candidate questions that would fill it. Tick them to track" on hover; after a thin divider, Question type ▾ (the dimension menu: Question type, Buying stage, Demand evidence, Topic and Customer role; the button shows the current one) and Source ▾; on the right, the amber search evidence chip (see the next point) and the Search candidates box. Once a candidate is ticked, the row turns into the select-all box, Track n and Remove.
  • Search evidence chip: when most of the market's tracked questions have no real search evidence, a small amber chip sits at the right of the toolbar, e.g. "⚠ 105 (85%) without real search evidence". Hover it for the full sentence ("Of the N tracked questions in <market> (brand ones aside), n (P%) have no real search evidence yet.") and what to do; when Search Console is not connected, the chip has a Connect Search Console link. Its ✕ closes it; it then stays hidden for that market until the share clearly rises or some time has passed. It counts one market's tracked questions without brand ones, so the number can differ from the total in Prompt Tracking.
  • One group row per kind, with its checkbox on the left edge: with Question type they are Discovery, Comparisons, Products & services, Customer pain points, Use cases, Industry trends and Brand. Beside the name are small chips: a signal-bars chip reading Enough data, Little data or No data yet (hover it for the period and number of answers counted, the 95% range and how far to trust the share), and on a thin question type Find this kind with a search icon, which looks for candidates for that kind only. On the right, each in its own column: a share such as Mentions you 13%, then the pills n tracked (the prompts of that kind you track in this market; paused ones and candidates aside), n answers and Candidates n; on a phone they sit on a line under the name. Grouped by Question type or Buying stage, n tracked is a link that opens Prompt Tracking filtered to that kind (or stage) and market.
  • Every kind starts folded; click any kind to open it. A kind with no candidates yet says so when opened and offers Find this kind; its checkbox is greyed out. An open kind lists only candidates not tracked yet; the tracked prompts are in Prompt Tracking. After a find, the kinds that got new candidates open by themselves.
  • Each candidate has a checkbox, the question, a source tag (Search Console · n impressions / Search suggestions / Related questions and searches / Forums (Reddit and others) / Your keywords / Inferred by AI / AI-generated) and "From the search: <original query>"; new ones are marked New. Hover over or click the source tag for Source details (the source, the original query, how it was written, how it was found). Hover over a candidate for ✎ (edit) and ✕ (remove).
  • Candidates are ticked when found, except those written by AI (AI-generated, Inferred by AI). The checkbox on a group row ticks the whole kind.
  • Once something is ticked, the toolbar turns into the select-all box, Track n and Remove. Tracking needs enough balance for the new prompts' first check (see below). Remove only takes candidates out of the list and never touches tracked prompts; Undo n brings them back. Each added prompt keeps where it came from.
  1. Demand evidence: with the dimension menu set to Demand evidence, the table has four rows: People search it (Search Console), Seen in search (Google autocomplete / People also ask), No search evidence (written by AI or chosen by you) and Brand searches. A note below says to judge visibility by the first two rows; prompts with no search evidence and brand prompts usually read higher.
  2. Topics and customer roles are grouped automatically the first time you open them. With the dimension menu set to Topic or Customer role, Edit groups and Regroup follow Source ▾ in the toolbar, to change the groups line by line or start again.
Prompt Library · Coverage and candidates
Question typePromptsMention youCandidates
Discovery · Enough data12 trackedMentions you 18%Candidates 6
Comparisons · Little data3 trackedMentions you 9%Candidates 4
Customer pain points · Find this kind2 tracked—
👆 Open a kind, tick candidates, then Track n

Tracked prompts (in Prompt Tracking)

The prompts you track are listed in Prompt Tracking, one per row; what you used to do on the My Prompts tab is done there:

  • Question type: the row's ⋯ → Question type: <current type> opens the seven kinds in place (Discovery / Comparisons / Products & services / Customer pain points / Use cases / Industry trends / Brand); pick one. It replaces the former manual Brand / Competitor / City / General classification.
  • Pause / resume: the row's ⋯ → Pause tracking / Resume tracking. A paused row carries a Paused tag.
  • Scan frequency: set it in the prompt's Evidence (Daily / Weekly / Monthly).
  • Delete: the row's ⋯ or the bulk bar; deleted prompts can be restored from the recycle bin for 30 days.
  • Fill one kind of question: in the Prompt Library's Coverage and candidates, click Find this kind beside the kind's name, or Find candidate questions (in place of the former per-group Generate prompts).
  • Add by hand: Add Prompt in the top bar, which takes many prompts at once (in place of the former input line at the end of each group).
  • Tags: prompt tags (Edit tags) are no longer offered; filter Prompt Tracking by Question type, Group or Market instead.
  • Every row in Prompt Tracking carries the same source tag (AI-generated for a prompt written by AI with no real search behind it); judge visibility by prompts with search evidence. For a question rewritten from a search, How it was written in Source details says so, and the question itself is not what the searcher typed. Older prompts added before sources were kept show Source not recorded; Citadex does not infer a source. Under Source details, Demand evidence lists the prompt's evidence by channel: evidence Citadex has not verified (for example a Search Console label set by an external app) is marked Unverified; evidence from another market (for example recorded before the prompt moved to this market) says Demand evidence from <market>. It does not represent this market. and does not count as this market's demand; the time window and impressions show when they were recorded. Channels are listed separately and never added together; Search Console impressions are the site's own impressions, not total market demand.
  • Demand heat: every row in Prompt Tracking and every candidate under Coverage and candidates shows Heat beside its source tag: five bars for the prompt's relative demand heat in its own market (level 1–5), or Heat Unknown when no demand evidence counts for this market. Hover or click to see the level, the channel used, the basis, the time range and country (Search Console), the confidence (High / Medium / Low), when it was read and the method version (heat_v1). Search Console is graded on five levels from the site's own impressions over the last 90 days in the market's country; Google suggestions, People also ask, related searches and forums only say that a search appeared, at most level 2; the level comes from the strongest single channel, and channels are never added together. Evidence from another market, unverified evidence and searches run by AI engines are not counted. Search Console impressions are the site's own impressions in Google Search and do not represent total market demand. When it shows Unknown, connect Search Console, or use Find candidate questions in the Prompt Library, to add demand evidence.

Quick Test (top bar)

Quick Test opens a dialog: type a prompt, choose models, click Run test, and see ✓ Mentioned / ✕ Not mentioned per engine, whether you were cited, and your rank. Use it to check one sentence; to track it, add it with Add Prompt in the top bar.

Import (top bar)

Import opens a dialog: the Google Search Console import at the top; below it, file upload: CSV or TXT, one prompt per line. For multi-column CSVs only the first column is read, a keyword / prompt header row is skipped, and duplicates are removed. Import first checks the list; take the suggestions one by one or with Use all suggestions, then click again to upload. The result shows how many were imported and how many duplicates were skipped.

The dialog works on the market currently selected on this page and names it at the top (for example "Market: United States"): the Search Console list shows the searches in that market's country, and every imported prompt (including those from a file) is tracked in that market. When the market changes, the list is read again for the new market's country and earlier picks are cleared.

Add Prompt (top bar)

The dark Add Prompt opens the Add Prompts dialog, headed “Paste the questions to track, one per line. Duplicates and prompts already tracked are skipped.” From top to bottom:

  1. Import from Search Console: pick from your site's Search Console queries; the ones you pick are added to the box below.
  2. The prompt box: one prompt per line. The line under it shows n recognized on the left and “Not sure what to add? Find candidates from real searches” on the right, which closes the dialog and takes you back to the Start from real searches. card at the top of the page (opened on another page, it reads “Not sure what to add? Find them from real searches in Prompt Library” and opens the Prompt Library).
  3. Market: only the project's existing markets, with the market chosen on this page first. To add prompts to a new market, choose + Add markets, the last item of the menu: the Add market dialog opens on top, and the market you add there is selected at once.

The dialog has no country, language, city or engine fields, and no type to pick: question types are assigned automatically. Track n first runs the check before adding; when it suggests a rewrite, or moving a prompt to the market in that prompt's language, take the suggestions one by one or with Use all suggestions, then click again to add. Tracked prompts go into Prompt Tracking and are checked, which uses credits; when the balance is too low, the dialog shows only the current balance and nothing is added.

Common tasks

Give a new market its prompts

  1. Click + Add markets beside the Market menu in the Prompt Library, choose the country, language, an optional city and the scan engines, then click Add market. Adding a market uses no credits. If the country and language are already a market of the project, the dialog says “<market> is already a market of this project, so there is nothing to add”, points to Markets & engines for changing its engines, and offers Switch to this market instead of adding it again.
  2. The Prompt Library switches to the new market, shows a green notice “Market <market> added. Candidate questions are being found from its real searches below; checks start once you track some.” and runs Find candidate questions for it at once; finding uses no credits.
  3. Tick the candidates you want in the kinds that open and click Track n; checks start once the prompts are tracked. If you already have a list of questions, you can also click Add Prompt, choose this market under Market and paste them.

Add real search queries from Search Console

  1. Connect Google Search Console under Integrations → Google first. Once it is connected, the AI assistant follows up once in the chat with your site's real questions that you don't track yet, and adds them when you confirm the card.
  2. Find candidate questions in the Prompt Library: once connected, it uses your site's own searches automatically; those candidates carry the Search Console · n impressions pill.
  3. To pick queries yourself, use Or pick a source → Import from Search Console and click Add to candidates n; to skip the candidates and import straight away, use Import in the top bar → Google Search Console.

Find and fill a coverage gap

  1. Open the Prompt Library and use the dimension menu in the Coverage and candidates toolbar (it reads Question type ▾ by default) to switch between Question type, Buying stage, Demand evidence, Topic and Customer role.
  2. Look for kinds marked Little data or No data yet.
  3. Open that kind, tick the candidates that fit and click Track n; on a question type with no candidates, click Find this kind beside its name, or find more with Or pick a source or Or expand a seed.

Manage markets

A market can only be removed when it has no prompts, and the primary market can’t be removed. When it can’t be removed, Go clear them opens Prompt Tracking filtered to that market, where you can move its prompts to another market or delete them. From the markets and engines settings you can also choose which engines scan each market.

Common questions

A payment was refunded or disputed. What happens to the credits? A refund, in full or in part, takes back the same share of every credit that payment brought: the purchased credits, their top-up bonus, the first-top-up bonus and the inviter's reward (for a plan payment, that period's plan credits). Credits still in the wallet are taken back; credits already used are recorded as a debt, and no credits can be used until top-ups have paid it. Credits that expired unused are not a debt. While a payment dispute is open, the same share is on hold and cannot be used; it is released if the payment stands and taken back if the payment is reversed. A refund that does not go through, or a dispute in which the payment stands, returns the credits (Credits returned in the credit history). Auto-recharge is switched off after a refund or dispute and cannot be switched back on while a dispute is open.

Credit use rose sharply. How do I see where it went? Mostly automatic re-checks: a prompt set to Weekly is re-checked once its last check is 7 days old (Monthly: 30 days). Open API keys & credits → Platform credits → Credit history: the <project> automatic re-checks row has Details with each engine's checks, credits and refunds for failed checks. When many automatic re-checks are due the next day, or the balance may not cover them, the bell reminds you a day ahead. To reduce usage, open a low-value prompt's Evidence in Prompt Tracking and set Scan frequency to Monthly; for prompts that stay at 0%, use Pause tracking in the row's ⋯.

I set a market to the wrong language. How do I change it? In Markets & engines (top bar), the language on each market's line is a dropdown: pick the new language, no need to delete and re-add the market. Existing prompts keep their wording; from the next check the AI engines are asked in the new language, and the market's engines adjust to it.

Is there a limit on prompts? No. No plan limits how many prompts you track. Each scan is billed in credits per prompt and engine, so remove low-value prompts or set them to Monthly.

"Not enough credits to add prompts"? Your balance doesn't cover the new prompts' first check, so nothing was added. Top up in the dialog and add them again, or add fewer.

Generated prompts aren’t relevant to the business. Find candidate questions looks in the market's real searches first; only when no real search fits are questions written by AI (tagged AI-generated). Connect Search Console so your site's own searches are used too. Tick candidates that don't fit and click Remove.

I can’t set Daily. The number of prompts on daily frequency is capped per plan, across the account: 50 on Free and Pro, 100 on Business. Set other prompts to Weekly or Monthly, or upgrade.

Troubleshooting

  • No valid prompts detected in the file.: check that the file has one prompt per line and is UTF-8.
  • Google Search Console isn’t connected: go to Integrations → Google Search Console and connect it.
  • Today’s searches are used up. Try again tomorrow.: Find candidate questions and the sources can be used a limited number of times a day; try again tomorrow.
  • Failed to add prompt. Please try again.: usually a network issue. Refresh and retry; prompts that were already added are not duplicated.

Was this article helpful?