Working with Prompts
Prompts are the questions users ask AI platforms that you want your brand to be visible for. ReachLLM runs your tracked prompts against each enabled AI platform on a schedule and analyzes every response, so your prompt list defines exactly what your visibility scores measure.
Where prompts come from
Section titled “Where prompts come from”Use Generate to prepare reviewed additional questions with the same business-fit and quality checks as onboarding. For ordinary accounts, it saves questions to Saved in the background, so you can review them before deciding what to test or track. Existing prompts remain. Generation has no execution-credit research fee and does not start a paid visibility analysis. In the Saved tab, Generate to Saved uses this same workflow.
Prompt explorer is an admin preview. Authorized ReachLLM platform admins can browse products and services, buyer needs, topics and question clusters, then choose reviewed suggestions to track. It is hidden from ordinary customer accounts and team admins. Its research and suggestion actions use no execution credits.
When you set up a brand, ReachLLM generates a starter set of prompts automatically from your brand analysis. These suggest questions your likely customers may ask AI assistants about your category, your products, and your competitors.
You are not limited to the generated set. You can:
- Add your own prompts for questions you know matter to your business.
- Edit any prompt to sharpen its wording.
- Archive prompts you no longer want to track. Archived prompts stop running and stop counting against your tracked-prompt limit.
Adding prompts
Section titled “Adding prompts”Open Your Prompts > Add prompts and enter one complete question per line, or upload a CSV. Saved is the default destination. Choose Active tracking if the questions should join future recurring runs. Blank lines and duplicates are skipped. Neither destination starts a visibility check automatically.
Your Prompts has four views:
| View | What belongs here | Runs on your tracking schedule? |
|---|---|---|
| Active | Questions you want to monitor repeatedly | Yes |
| Saved | Ideas and proposed questions you want to review or test first | No |
| Archived | Questions you have set aside, with earlier answers retained | No |
| Tests | Saved one-off executions and their actual answers | No |
Move Active questions to Saved to stop recurring checks while keeping them available. Archive sets a question aside. Select Saved or Archived questions and choose Start tracking when you want them in Active. Only Active prompts use your tracked-prompt allowance. These actions preserve historical results and do not start paid checks.
The current Brand Visibility view combines the latest successful tracking answers for your Active questions. Newly activated questions without tracking answers remain pending until a normal visibility run checks them. Moving a question out of Active removes its contribution immediately. Pending or failed checks do not erase earlier successful tracking answers. Refresh Now requests fresh answers across your Active set. Historical runs keep their original questions and answers.
Test questions without daily tracking
Section titled “Test questions without daily tracking”Select questions in Active, Saved or Archived and choose Run selected, or use Run once on a Saved or Archived row. ReachLLM first shows the exact questions, enabled-platform logos and total execution-credit cost. Confirm Run once to start that quoted test in the cloud. You can leave the page and return through Tests while it continues.
One-off testing uses the same per-prompt execution-credit rate as normal Brand Visibility, covering the brand owner's enabled platforms together. There is no added research fee. The final receipt shows credits used and returned. A completed negative answer can still use credits; failed or untriggered checks are not evidence that the brand was absent.
Results distinguish a brand mention from a citation to its website. Filter Matches or All, and Branded or Unbranded, then open compact evidence for the actual answer, platform, date and sources. A branded question does not become a non-branded discovery. Unknown or unresolved checks are identified separately.
Tests preserve the quoted questions, brand identity, attributes and platform/model settings. Editing a question or its labels later does not rewrite an earlier test. Test history compares the same saved question set and settings. Tests do not add questions to Active, change the schedule or replace normal visibility metrics and trends. Viewing saved results does not run them again.
Your tracked-prompt allowance is checked across the workspace before the batch is saved. If the batch would exceed the allowance, ReachLLM shows the current usage and limit without clearing your draft.
A tracked-prompt warning counts active saved prompts, including other projects in a pooled workspace. It does not show execution-credit spending. Active prompts stay in your inventory across billing cycles, so waiting for a credit reset does not free prompt capacity. Archive or remove prompts you no longer need to track, or increase the workspace allowance.
Discover prompts where your brand appears
Section titled “Discover prompts where your brand appears”Hover over, focus or click the information icon beside Discover prompts for a short explanation. Click to keep the help open while you read it; click again, press Escape or click elsewhere to close it. The help stays above navigation and fits within your screen.
Open Your Prompts for the brand you want to review, then expand Discover prompts. The information icon explains the workflow. Find now prepares up to 25 relevant conversational prompts, then checks each exact question across all your enabled AI platforms using the same provider, model and location settings as Brand Visibility. Questions can be concise or include a short grounded buyer scenario and a request to compare, evaluate or solve a problem. Discovery prioritizes different customer needs instead of filling the list with keyword variations. If the available context supports fewer useful questions, it checks fewer and returns the unused reservation.
Discovery starts in the background. You can navigate elsewhere or close the page while it runs. Return to Your Prompts > Discover prompts to see saved progress, completed platform checks and results. A running scan is not an empty result. You cannot start a second run for the same brand while one is active. Temporary problems while preparing prompts get one automatic recovery attempt before platform checks begin. Incomplete output is never used as a finished prompt list. If preparation still cannot finish, no platform checks run, all reserved credits are returned, and your saved discoveries remain available. After that fully refunded preparation failure, you can choose Find now again immediately. Active scans, checks that already ran, and the monthly scan limit keep their existing protections.
A new discovery is saved only after a checked answer mentions your brand or cites your website. Once saved, its positive answer and source evidence stay available across future runs until you explicitly archive it. A later negative answer, incomplete check, older check date or change to enabled platforms does not erase that evidence. Each finding shows its recorded date. Earlier evidence is labeled as an earlier check and does not claim that the brand still appears today. The compact discovery toolbar shows the enabled platforms with their logos, Find now, and the maximum credit cost. Matching-platform logos identify where each result was found. Open Evidence to see the checked date, switch between platform answers, read a compact matching passage, and inspect the source links. The preview includes the explanation beneath a matching brand heading. Links to your website appear first, with up to three sources shown initially. Expand the source list to see every saved link. Read full answer expands the response. A written brand mention and a website citation are shown separately. Source links are preserved in the saved answer. If a platform does not provide a source title, the page address is shown. A new run can recheck questions whose earlier citation evidence was incomplete.
Generic product and service labels, such as a GEO audit, do not establish a brand mention. Discovery checks actual brand names and explicit aliases separately from the labels used to exclude branded product questions.
Why this prompt? in the evidence view explains whether the question was recorded or suggested from a source topic, with a supporting page when available. Generated wording is identified separately from recorded wording. Rewriting a keyword does not transfer its older citation or establish how many people ask the new question.
View last scan shows the exact questions tested and the number of completed or incomplete platform checks, even when nothing qualified for review. A negative check is not an opportunity to add to tracking, and the scan does not measure how often people ask a question. Scans set aside for question quality are labeled separately.
Questions containing your saved brand names, alternate names, website domain or distinct own product names are excluded, including spaced spellings and model numbers. Generic offering labels such as product analytics or session replay remain valid discovery topics. Links to third-party app and media listings on shared catalog hosts do not count as citations to the parent brand's website.
Choose Add to tracking for a prompt you want to keep monitoring. It uses your existing tracked-prompt allowance and joins your next visibility analysis. Discovery does not automatically add prompts, create a tracking analysis, or change your visibility scores.
Keeping and archiving discoveries
Section titled “Keeping and archiving discoveries”Use Archive on a discovery when you no longer want it in the main review list. Open Archived to inspect it or choose Restore. Archiving preserves the question, its positive answer and sources. It also keeps the question out of later discovery runs until restored. These actions do not spend discovery credits, change your background schedule or remove an existing tracked prompt. View-only team members can inspect evidence but cannot archive or restore it.
Saved discoveries are separate from the latest scan. View last scan describes the most recent attempt, including negatives and incomplete checks, while the saved list keeps earlier positive findings. Use Load more when older saved discoveries are available. Adding a discovery to tracking marks it Tracked and retains its evidence.
The platform logos beside Find now and in its cost confirmation come from the brand owner's enabled settings, matching Brand Visibility. Only those platforms are checked in a new run. Saved evidence preserves the platforms that actually checked it at the time. A previously used platform can appear as an earlier check even when it is no longer enabled for new runs.
Credits and background discovery
Section titled “Credits and background discovery”Before a manual run, ReachLLM shows the enabled platforms and maximum execution-credit cost. Discovery uses the same 20 execution credits per prompt as a manual Brand Visibility refresh, up to 500 credits for 25 prompts. Research and question preparation have no extra credit charge. The per-prompt cost covers all enabled platforms together and applies when at least one platform returns a completed answer. A completed check uses credits even when the answer has no brand match.
The maximum is reserved once when the run starts and settled after all checks finish. The final cost and returned unused credits appear in View last scan. Unused reserved credits are returned, including prompts where every platform failed. A preparation-only run or a run with no completed platform answers costs zero execution credits. Historical scan receipts keep their original charges and dates.
Background discovery starts Off. Choose Weekly or Daily, save, and review the same maximum cost per run before enabling it. The next run date appears in the section. Background runs use the same credit policy as Find now, save matching prompts for review, and never add them to tracking automatically. Switch the schedule off anytime. Changes to platforms, brand scope or a higher budget require review before discovery resumes. Removing the research fee preserves an existing matching 25-prompt schedule and its next run date at the lower cost. Older schedules approved for five prompts still need approval of the 25-prompt maximum. Scheduled discovery uses the same durable background checks; a background run does not imply every provider offers a discounted Batch API.
Each brand can run up to 32 scans per calendar month, normally with at least one hour between attempts. The one-hour wait does not apply to a fully refunded preparation failure where no platform checks ran. Manual, background, and agent scans share the monthly limit. Adding a saved suggestion does not spend discovery credits.
How questions are found and checked
Section titled “How questions are found and checked”Previously recorded citations can supply candidate questions or topic seeds. Short search phrases are converted into separate natural questions, which receive their own fresh checks. A broader bounded seed pool is filtered for the selected business and varied customer intents, so unrelated parent-company articles and duplicate wording do not take every slot. ReachLLM can also narrow the search using saved website pages, audit topics, and your brand context. When connected AI crawler traffic is available, pages visited by crawlers can help prioritize candidate topics. System files and crawler probe URLs are skipped, including obsolete questions inferred from those files in earlier scans. Page activity does not reveal the original prompt or prove that a page was cited. All candidates must pass a fresh live answer check before appearing in the review list.
The live checks use the enabled platforms' API or search surfaces. Results can vary between accounts, locations, models, and runs, so a match does not promise that every consumer chat session will return the same answer. Discovery does not reveal private user conversations or every prompt ever asked. No matches means no qualifying current answer was found in this run, rather than proving your brand is never cited.
Historical records alone are no longer shown as verified opportunities. Saved findings show when their positive answers were checked. An unsuccessful or incomplete recheck keeps that earlier evidence available and labels it as an earlier check. Only archiving hides the discovery from your saved list. Use the exact prompt and cited page to investigate what is working, then track the prompt to measure whether it continues. A citation proves the saved answer linked the page; it does not prove why the AI selected it.
Reading prompt percentages
Section titled “Reading prompt percentages”Your Prompts opens on Latest 5 runs. Each row pools the successful answers to that question from the brand's five most recent eligible visibility runs, or all available runs if there are fewer than five. Each platform can contribute one answer per question in each run. A tracking run in progress can contribute answers that have already finished. One-off tests and Prompt explorer checks are separate.
| Percentage | How to read it |
|---|---|
| Visibility | The share of successful answers that mention your brand. |
| Citations | The share of the same answers that link to your configured website or a subdomain. Multiple links in one answer count once. |
A run that skipped the question contributes no answers to its row. Failed checks and Google searches that returned no AI Overview also contribute no answers. A successful answer with no mention counts in the denominator. For example, suppose one run produced one answer to your question and it mentioned you. Another run produced three answers and one mentioned you. Across those runs, 2 of 4 answers mentioned you (50% visibility). If three of the four answers linked to your site, the same row shows 75% citations.
Hover or keyboard-focus a percentage to see the counts and contributing runs. A dash indicates that the row has no usable answers, or that the brand needs a supported website to calculate citations. Open the prompt to inspect its platform answers and cited sources.
Choose Include 5 older runs in the table or in a prompt's details to add up to five earlier runs. The table, open prompt, group scores, and CSV export use the same set of runs. You can repeat this while older runs are available. Choose Back to latest 5 runs to return to the default. Reviewing older runs uses saved answers and spends no execution credits.
Adjusting the table
Section titled “Adjusting the table”On desktop, drag the thin divider at the right edge of any column heading to make that column narrower or wider. This includes the Prompt column. Your widths are saved in this browser. If the table becomes wider than the available space, scroll it horizontally.
You can also focus a divider and use the arrow keys, or hold Shift for larger changes. Press Home or double-click the divider to reset that column. Reset column widths restores the full default layout. On smaller screens, each prompt keeps a compact stacked layout with both percentages.
Organizing prompts
Section titled “Organizing prompts”A well-organized prompt list makes your visibility results much easier to read.
Audience and question attributes
Section titled “Audience and question attributes”Select questions and choose Attributes to save country, region, city, main topic, subtopic, intent, associated products and audience. These labels help filter and compare groups of questions. The audience location describes the intended buyer; it does not change the AI platform's native search location. Saved CSV imports can also include Country, Region, City, Intent and Product alongside Topic and Prompt.
The agent can propose questions with these attributes. Agent drafts remain drafts; reviewed generation uses the shared quality checks. Neither a useful label nor a generated wording proves that people submitted that exact AI prompt.
View visibility for a selected set
Section titled “View visibility for a selected set”In Active, select the questions you want to compare and choose View visibility. The selected set scopes visibility cards, response tables, source analytics and trends. For example, selecting ten questions measures those ten, rather than the whole brand. Existing branded/unbranded, product and topic filters can narrow the set further. An unavailable or empty selection stays empty rather than reverting to all questions.
If a grounded sentiment assessment for that exact selection is unavailable, sentiment is shown as unavailable. Clear the selection to read the separately labeled full-analysis Brand Perception report.
Topic tags
Section titled “Topic tags”Group prompts by theme with topic tags. Tags let you filter results and power the Topic-wise Visibility heat map in Brand Visibility, so you can compare your cumulative mention rate with competitors for every topic.
Open Knowledge Base > Topics to add, rename, or remove a topic and assign prompts to it. ReachLLM confirms each saved change. If saving a new topic fails, the previous topic list stays in place and the input keeps your draft so you can try again.
Each prompt can belong to more than one topic. When it does, its tested AI answers contribute to every assigned topic. Topics with no analyzed prompts stay out of the heat map, and intent stages remain separate from topics.
When you export generated search prompts from Knowledge Base, the CSV filename uses the date the saved prompt set belongs to. Downloading an older saved set does not replace that date with today's date.
Branded vs unbranded
Section titled “Branded vs unbranded”Each prompt is classified as branded (it names your brand) or unbranded (a discovery question that does not). ReachLLM sets this automatically when prompts are generated, and you can change it: click the Branded or Unbranded label on any prompt to flip it. This classification powers the Any / Branded only / Unbranded only filter in Brand Visibility, so getting it right keeps your branded and unbranded views accurate. See AI visibility tracking for how the filter works.
Intent stages
Section titled “Intent stages”Assign each prompt an intent stage that matches where the asker is in their journey:
| Stage | What the asker wants |
|---|---|
| Learn | Understand a problem or category |
| Consider | Compare options and shortlist solutions |
| Purchase | Pick a specific product or vendor |
Tracking visibility by stage shows whether you appear early in research, late at the decision, or both.
Linking prompts to products
Section titled “Linking prompts to products”You can link prompts to specific products, so you can see how each product line performs in AI answers rather than only viewing brand-level results.
Bulk operations
Section titled “Bulk operations”You do not have to organize prompts one at a time. Bulk tagging and bulk product assignment are supported, so you can select many prompts and apply tags or product links in one action.
Tracked-prompt limits
Section titled “Tracked-prompt limits”Each plan includes capacity for active tracked prompts:
| Plan | Active tracked prompts |
|---|---|
| Pro | 100 |
| Scale | 500 (pooled across up to 10 brand projects) |
| Enterprise Agencies | 1,000 (pooled across unlimited brand projects) |
| Growth | 500 for one managed brand project |
| Enterprise | Custom |
See Plans and pricing for full plan details. To stay within your limit, archive low-value prompts and keep the list focused on questions that drive real buying decisions.
Selected-plan trials and complimentary access use the same tracked-prompt allowance as the displayed plan. When the allowance is full, ReachLLM prevents another prompt from being added and offers an upgrade through Account > Billing. Existing prompts and saved results remain available.
Some custom agreements include a higher account-level tracked-prompt allowance. Billing, prompt creation, manual reruns, scheduled tracking, and Agent-triggered reruns all use that same effective allowance. Pooled plans count active prompts across the whole workspace.
Scheduled prompts via the agent
Section titled “Scheduled prompts via the agent”Beyond the recurring visibility runs, you can have the ReachLLM agent run a task on a schedule: daily, weekly, monthly, or once. Set these up and manage them in the agent's Scheduled panel. This is useful for recurring checks or reports built around your prompts, with results delivered without you having to ask each time.
Learn more about what the agent can do in The ReachLLM agent.
Next steps
Section titled “Next steps”- AI visibility tracking: how your prompts turn into scheduled runs and analyzed answers.
- Understanding the scores: how Visibility Score, Share of Voice, and rank are calculated from prompt results.
- The ReachLLM agent: generate prompts, analyze performance, and schedule recurring tasks.
