AI Visibility Tracking

AI visibility tracking shows you how AI assistants talk about your brand. ReachLLM runs your tracked prompts against each enabled AI platform on a schedule, analyzes every answer, and turns the results into scores, trends, and source intelligence.

How tracking works

You track prompts. Prompts are the questions users ask AI platforms that you want to be visible for. They are generated automatically from your brand analysis, and you can add, edit, or archive your own. See Working with prompts.
ReachLLM runs them on a schedule. Each tracked prompt runs against every enabled AI platform daily or weekly, depending on your settings.
Every answer is analyzed. ReachLLM checks each response for mentions of your brand and your competitors, classifies the sentiment of each mention as positive, neutral, or negative, and records which sources the platform cited, including whether it cited your own domain.
Results become metrics. The analysis produces your Visibility Score, Share of Voice, Average Rank, sentiment distribution, and citation rate. For the exact math behind each metric, see Understanding the scores.

You can also start a manual Refresh Now run from Brand Visibility. ReachLLM shows the prompt count and execution-credit estimate before the run starts. When you confirm, the progress screen appears immediately while setup continues, and the start action stays locked. The browser and server share the same active-run protection, so a second click, another tab, or a scheduled refresh cannot create a duplicate paid analysis for the same brand. If a run is already active, ReachLLM reuses that run instead. Every manual refresh asks each enabled provider for a new response, even when the same prompts ran earlier that day. Your previous results stay available until the new ones are ready. If you hit Cancel, the run stops right away and nothing from it is saved, so you can safely fix your prompts and refresh again without an old run reappearing later. When the ReachLLM agent starts a Brand Visibility run for you, it uses the same execution-credit charge and stops before starting if your monthly allowance and top-up balance are exhausted. Auto-refresh can be daily or weekly, and you can turn it off from the same settings dialog.

Scheduled runs use the exact saved Next run time, including the local weekday chosen when auto-refresh was enabled. The daily and weekly schedulers check that timestamp each hour, so a local midnight that falls on the previous UTC day is not missed. An overlapping manual or scheduled run is skipped without another usage charge, and the next scheduled time advances normally. Scheduled runs enqueue the complete active prompt set before ReachLLM treats the refresh as started. When processing finishes, the completed analysis is saved to history and the auto-refresh record is marked completed. If one or more providers need another check after the prompts finish, the progress screen names those providers and shows how many responses remain. Completed prompts are shown separately, so a finished prompt count is not mistaken for a finished report. If all providers fail, or a scheduled run stops reporting progress, ReachLLM records the failure and releases the run instead of leaving auto-refresh stuck on processing. If a run is skipped or fails, Brand Visibility shows Auto-refresh needs attention with the saved reason instead of showing a healthy next-run countdown.

Scheduled tracking uses the workspace owner's live entitlement. This includes assigned teammates and Paddle-managed Enterprise Agencies workspaces. If an old saved run says billing was unavailable but the current workspace plan is active, Brand Visibility clears that outdated billing warning while preserving real quota and provider failure notices.

By default, Brand Visibility shows the latest successful evidence for your current active prompts. Adding prompts runs only the additions, then combines their completed answers with the saved answers for the rest of your prompt set. Archiving prompts removes their contribution immediately. Visibility Score, Share of Voice, rank, citation rates, topic breakdowns, sources, Query Fanout, and Detailed Responses all use that same current evidence. Updating these totals does not start another paid AI run.

An in-progress or failed check does not replace an earlier successful answer for the same prompt and platform. Prompts without a completed answer remain visible as pending coverage rather than counting as a successful answer with no mention. Sentiment remains tied to its saved full analysis and shows its scope separately. Historical runs and trend points retain the prompts and answers recorded for those runs; opening history does not apply today's prompt additions or removals.

Platforms

Pro, Scale, and Enterprise Agencies plans include four AI models: ChatGPT, Google AI Overviews, Perplexity, and Gemini.

For shared brands, available platforms and model choices come from the brand owner. Every assigned teammate sees the same enabled platforms, including Claude when the owner has it enabled, and the same results when viewing the same prompts, dates, and filters. A run started by a teammate uses the owner's configuration, just like a scheduled run.

Brand Visibility keeps every measurement tied to its named source. ChatGPT calls OpenAI directly, Gemini calls Google directly, Claude calls Anthropic directly when it is enabled, and Perplexity calls Perplexity directly. Google AI Overview checks live Google results through SerpAPI. ReachLLM does not substitute OpenRouter or another aggregator when one of these providers is unavailable. That provider is retried and then shown as failed instead of being relabeled.

Temporary connection errors, rate limits and empty answers are retried automatically within a bounded recovery window. An empty answer is never treated as successful just because it includes source links, and a retry bypasses unusable cached answers so it can contact the original provider again. Repeated failures pause the affected provider for a 15-minute cooldown before queued work resumes. A provider's content restriction affects that response and does not count as a service outage. Persistent outages, invalid credentials and exhausted provider billing still need operator attention; recovery does not create new API keys, add credit or retry forever.

ReachLLM sends its operations team an email when provider access fails and a recovery confirmation after the next fresh, successful direct check verifies access after the latest failure. Recovery emails go to the same recipients as outage alerts and are deduplicated for each incident. Cached answers and declined prompts cannot confirm recovery, and an active provider cooldown must finish first. A successful initial check, automatic retry or manual retry can confirm recovery. If email delivery fails, the next fresh successful check can retry the notification. This confirms provider access; it does not change previously failed results. A new failure after recovery starts a new alert incident, even within the earlier alert's cooldown.

ChatGPT uses GPT-5.6 Sol Medium by default. Accounts with AI model controls can choose another GPT-5.6 Sol reasoning level in Account settings > Profile. If an older saved model is no longer available, Brand Visibility safely returns to GPT-5.6 Sol Medium so the ChatGPT check can continue.

Gemini uses Gemini 3.7 Flash by default. Accounts with AI model controls can choose Gemini 3.7 Flash, Gemini 3.6 Flash, or Gemini 3.1 Pro in Account settings > Profile. Retired or unknown saved model names safely move to a current choice before a manual, scheduled, or retried check runs. Every Gemini Brand Visibility measurement uses Google Search grounding so its answer can draw on current web evidence.

For a free AI Visibility Report, delivery is stricter: all 50 prompts must complete successfully for ChatGPT, Google AI Overview, Perplexity, and Gemini before the report can be published or emailed. A successful Google check where no AI Overview appears is recorded as a valid no-trigger result. If an isolated provider response remains incomplete after its normal retries, ReachLLM automatically retries the report in the background with the same verified 50 prompts. Failed recovery runs advance through the scheduled retry attempts, so a temporary provider problem does not leave the report stuck after one retry. The verified prompt inventory and completed analysis become the customer's read-only account workspace at publication time. Email delivery is a separate notification and cannot block that access. A customer who signs in early sees report preparation status, then opens the same saved results in Brand Visibility without repeating brand setup or choosing a plan first.

Free report accuracy checks

Your report retains the competitors you selected, including those with no mentions or incomplete verification. Each entry shows its comparison status. ReachLLM also checks official websites for the closest alternatives that serve the same buyer need, then measures them against the saved AI answers. These alternatives are identified separately from your selections. Comparative percentages require at least three verified competitors with actual appearances. If evidence is thinner, the competitor list remains visible and your own visibility score is unaffected.

Free report research starts with the exact website you submit, using cited official-site evidence to identify the business, services, and audience. ReachLLM then checks all 50 proposed questions against that identity. It does not guess the category from a short brand name or substitute a similarly named company. If the identity or question review cannot be completed, the report waits for investigation instead of publishing unchecked results.

Alternate names should be genuine spellings, abbreviations, or names of the business. In free reports, unrelated aliases and category words such as "HVAC" or "preschool" cannot establish a brand mention on their own. The answer must also identify the actual business. A link to your website without a visible brand name can still count as a citation, but is not automatically a mention.

Perception is a separate assessment of sampled, recorded AI answers. New free report perception summaries use attributable quotes checked against those answers, with the platform and supporting evidence retained. Statements about a different business are separated from company claims. These are observations of what an AI platform said, not independently verified facts about your company. Missing or unusable evidence leaves the perception score unavailable; it is not converted into a negative score.

Sentiment scores use complete sampled answers rather than votes from selected positive sentences. Mixed answers retain their caveats, and an explicit provider rating is shown separately from the sentiment index. Topics are drawn from the business and its supporting quotes, rather than a fixed software-category list. The page and newly generated reports show partial or unavailable evidence honestly, even when a platform returned visibility answers. Report slides use compact citation labels to keep long source addresses from crowding out the assessment; supporting evidence retains its original text and source addresses. Older saved sentiment is refreshed from the recorded answers when needed, without starting a new visibility run. Recovery retries missing assessments while preserving usable provider results from the same analysis. See Understanding the scores for the calculation and limits.

Before publication or a report email, ReachLLM checks that the completed analysis, its processing record, and the saved report refer to the same reviewed website and exact 50-question inventory. Older shared reports are saved snapshots and do not silently change when these checks or matching rules improve. See Reports and audits if a report needs correction.

You can extend coverage with model add-ons, billed per month:

Add-on	Pro	Scale	Enterprise Agencies
Google AI Mode	$50	$250	$500
Microsoft Copilot	$50	$250	$500
Claude	$200	$1,000	$2,000

See Plans and pricing for full plan details.

The results tabs

The Brand Visibility page organizes results into five tabs:

Overview: your headline metrics across platforms, including Visibility Score, Share of Voice, topic-wise visibility, and rank. Direct competitors from Knowledge Base are the default comparison group, with an All tracked option.
Sources & Opportunities: source analytics is the default view, including the domains and pages AI platforms cite, content-type breakdowns, and source history. Open the optional Gap Analysis view when you want prompt-level gaps against the selected comparison group.
Sentiment: AI-evaluated perception of your brand, including a 0 to 100 score, positive, neutral, and concern percentages, supporting evidence, and a Direct-competitor head-to-head based on independent mention rates.
Query Fanout: the subqueries an AI platform generates from each tracked prompt while building its answer, including how often each subquery appears across analyses.
Responses: the answers from each platform with their headings, lists, inline links, and clickable citations preserved. A linked brand label remains visible and counts as a mention, while a source link that does not name the brand remains a citation rather than a mention.

CSV downloads use the date of the analysis run shown on screen. If you select an older point in the Source Timeline, exports from Sources & Opportunities, selected domains, and Detailed Responses keep that historical run date in the filename instead of using the day you downloaded the file.

Topic-wise visibility

The Overview tab groups the analyzed prompt set by the topics assigned in Knowledge Base > Topics. Each column combines every tested AI answer for the prompts in that topic, then shows how often your brand or a competitor appeared. A prompt assigned to more than one topic contributes to each of those topics.

The heat map includes every topic that has an analyzed prompt in the current view. It leaves out intent labels such as Learn, Consider, and Purchase, as well as topics with no matching answers. Darker cells mean a higher mention rate. Hover, focus, or tap a cell to see the exact company appearances, tested answers, and prompt count behind the percentage.

Each new analysis saves its topic assignments with the run, so later topic edits do not rewrite that run's historical breakdown. Analyses created before Topic-wise Visibility use the current topic organization because they do not contain a saved topic snapshot.

Prompt and provider filters apply before the heat map is calculated. The Direct or All tracked comparison control changes which competitor rows appear. Product Line Visibility remains available when prompts are also linked to products, while Topic-wise Visibility works for every organized prompt set.

Exporting cited sources

In Sources, click Select, choose one or more domains, then click Export CSV beside the selection controls. The file includes one row for each individual cited URL in those domains, including URLs outside the currently loaded portion of the list. Use the main Export CSV button to export all sources in the current filtered view.

Each row includes the source title, URL, category, AI providers, citation count, citation percentage, original URL, and the prompts that cited it. Domain citation totals, domain percentage, unique URL count, and domain providers are repeated alongside each URL so you can compare pages within a domain or across competitors.

Exports follow the current source filters. Citation percentages use all citations in the filtered source view, matching the percentages shown on screen. Selecting a single domain does not change that denominator. For example, a URL cited twice out of 2,000 filtered citations exports as 2 citations and 0.1%. Domain totals are repeated for reference, so do not sum those columns across URL rows.

Branded vs unbranded prompts

Every prompt is either branded or unbranded:

Branded prompts name your brand directly, for example "Is Acme good for enterprise support?". Because the prompt forces the AI to talk about your brand, branded visibility is usually very high (often 80% or more). If it drops below that, the AI either does not know your brand well or is confusing it with a similarly named entity.
Unbranded prompts are discovery questions that do not mention you, for example "best help desk software for startups". Unbranded visibility is the real measure of how discoverable you are when people are not already searching for you by name.

Open Filter in Brand Visibility and, under Filter by Prompts, use the All / Branded / Unbranded quick-select to narrow the prompt set by type. Overview, Sources, Query Fanout, and Responses use the selected prompt set. Their scores, share of voice, rankings, and sources recompute for the prompts you selected. Sentiment is measured across the full analysis and shows a note when any page filter is active.

You can set each prompt's branded or unbranded status yourself on the Working with prompts page. When you generate a report while a branded or unbranded filter is active, the report is scoped to those prompts and shows a matching badge. The ReachLLM agent can apply the same filter: ask it for "unbranded visibility" or "a branded-only report".

Trends over time

Trends charts show how your score, share of voice, and rank change over time, per platform and per brand versus competitors. The platform view labels the union rate as Prompt coverage: the percentage of tracked prompts where at least one model mentions or cites you. This is different from Visibility Score, which averages the selected platforms' individual mention rates. Use the trends to confirm that your optimization work (content, GEO fixes, outreach) is moving the numbers.

Managing competitors

Competitors are managed from one Knowledge Base workspace. Brand Visibility uses that workspace as the source of truth:

Direct competitors are explicitly tracked peers.
Indirect competitors are adjacent alternatives that may compete for the same buyer attention.
SERP competitors appear organically in AI search results and can be promoted into the tracked comparison set.
Other companies stay visible for review, but are not included in tracked comparison metrics until you promote them.
Multi-select and batch actions let you change several competitors to direct, indirect, SERP, or remove them at once.
Drag and drop moves companies between the tracked competitor table and Other Companies review list.
Duplicate review helps you merge or ignore likely duplicates, including companies with matching domains or aliases.
Own-brand alternate names cover the ways your brand may appear, such as a one-word spelling, two-word spelling, abbreviation, or common misspelling. Add or remove them from Knowledge Base > Profile. ReachLLM confirms the save, then recalculates current and historical visibility from the stored responses so every view uses the same names.
Competitor aliases cover alternate spellings for tracked competitors so competitor mentions are matched correctly.
Remove wrong mentions. If a response was matched to the wrong brand, you can remove that mention and ReachLLM recomputes your metrics from the stored responses.

The Profile editor shows when a change is saving and confirms when it is complete. If a save fails, the previous value is restored and the editor asks you to try again. Alternative-name changes are not marked complete until the brand record is updated and stored visibility history has been recalculated.

Next steps

Understanding the scores: the exact definition of each metric.
Working with prompts: tags, intent stages, and product links.
GEO Audit: find and fix the on-page issues that hold visibility back.