> For the complete documentation index, see [llms.txt](https://boundaryai.gitbook.io/boundaryai-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://boundaryai.gitbook.io/boundaryai-docs/bringing-in-your-feedback/social-listening.md). # Social Listening **Agents** let BAI Analytics go out and collect feedback from the open web on your behalf (public reviews, app-store and employer reviews, social posts, news mentions, video comments) and bring everything back into the same Feedback Groups your surveys and connector data live in. One agent can watch many platforms at once, and the data is normalised into the same analytical model as everything else, so themes and sentiment span every source. This page covers agents end-to-end: which platforms are supported, how to create one, how the intensity tiers and credits work, how BAI Analytics keeps irrelevant content out, how to reply to what you collect, and how scraped data shows up in your dashboards. For other ways to bring feedback in, see [Creating a Survey](/boundaryai-docs/bringing-in-your-feedback/surveys/editor.md), [Uploading an Existing Dataset](/boundaryai-docs/bringing-in-your-feedback/uploading-an-existing-dataset.md), and [Connecting your tools](/boundaryai-docs/bringing-in-your-feedback/connect-to-your-existing-systems.md). *** ### When to use an agent Use an agent when: * You want to **track a brand or product** across review sites (Google Reviews, Trustpilot, TripAdvisor, Avis Vérifiés) without manually exporting reviews on a schedule. * You want **app reviews** from the Apple App Store and Google Play, or **product reviews** from Amazon, in the same place as the rest of your feedback. * You want **employer reviews** from Glassdoor and Indeed to understand how staff and candidates describe you. * You want **social listening** on your own channels or specific accounts (X / Twitter, Reddit, Instagram, LinkedIn, Facebook, YouTube, TikTok). * You want to monitor **press and news mentions** of a topic, product, or competitor (Google News). * You want public feedback to flow into the same Feedback Group as your surveys and support tickets, so themes can be compared across channels. If what you want to know is **how AI assistants describe your brand** (ChatGPT, Claude, Gemini, Perplexity, and others), that's a separate source type built on the same wizard: see [AI Visibility](/boundaryai-docs/bringing-in-your-feedback/ai-visibility.md). If the data already lives in a tool you control (a help-desk, a Jira project, a CRM), use a Connector instead. Connectors authenticate as you and pull native objects directly. Agents go after public data via web automation. *** ### Access and credits Web & Social listening is an add-on enabled per organisation by the BAI Analytics team. If it is not enabled for yours, the Agents page shows **Feature not enabled** with a note to contact your BAI Analytics representative. Agents are metered separately from the rest of the platform: * **Web & Social Media Credits** are a dedicated balance (independent of the Usage Allowance that runs surveys and analyses). You see the remaining balance in the Integrations Hub, under *Web & Social Media Credits*. * Each run **deducts credits up front** based on the intensity tier you choose (see below). You are only charged for data actually collected: if a run returns fewer items than estimated, the difference is refunded. * If a run **fails** before producing data, credits are **refunded automatically**. * Your organisation may have a **monthly credit allocation** set up by the BAI Analytics team. You can also **add credits** in fixed packages (25 / 50 / 100 / 250 / 500 / 1,000 credits, 1 credit = 1 USD) via secure Stripe checkout from the same section. * A full **credit history** is available from the same section, with one row per top-up, run, and refund. Balance is checked before every run. If you do not have enough credits, the run does not start and no charge is made. *** ### Supported sources Seventeen social-listening platforms are supported today, grouped by what they're best at: #### Review sites | Platform | Identifier | Notes | | ------------------- | ----------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Google Reviews** | Google Maps URL or place name | Star ratings preserved; auto-detects the place if you give it a business name. | | **Trustpilot** | Company review URL | Country subdomains supported (`trustpilot.com`, `.de`, `.co.uk`, etc.) so multi-region brands can be tracked under one agent. | | **TripAdvisor** | Listing URL | Hotel, restaurant, attraction, etc. | | **Avis Vérifiés** | Company review-page URL | French verified-reviews platform (avis-verifies.com). | | **Apple App Store** | App listing URL | `apps.apple.com/.../app/.../id...`. Apple caps reviews per country, so higher intensities pull the same app from several country storefronts (and languages). | | **Google Play** | App page URL | `play.google.com/store/apps/details?id=...`. Developer replies are captured alongside the review. | | **Amazon** | Product page URL | Any marketplace (`amazon.com`, `.fr`, `.de`, `.co.jp`, etc.). Long URLs with tracking parameters are accepted and tidied to the product's canonical link. | | **Glassdoor** | Company reviews URL | Any country domain. Reviews are anonymous; the reviewer's job title is shown as the author. Employer responses are captured. | | **Indeed** | Company reviews URL | `indeed.com/cmp//reviews` and country variants. Reviews are anonymous; the reviewer's job title is shown as the author. | #### Social platforms | Platform | Identifier | Captures | | --------------- | -------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------ | | **X (Twitter)** | Hashtags, account handles, search terms | Posts matching the filters. | | **Reddit** | Subreddits (+ optional keyword filters) | Posts and comments; configurable sort order. | | **Instagram** | Profile, post, or reel URL | Posts and comments. | | **Facebook** | Page or post URL | Posts and comments. | | **LinkedIn** | Profile, company, school, or post URL | Posts and comments. | | **YouTube** | Video URL, or a channel URL (up to 10 per agent) | Comments, with exact posted dates. A channel URL pulls comments from every video the channel published within your chosen timeframe. | | **TikTok** | Account handle, hashtag, video URL, or search term | Videos and their comments. | #### News | Platform | Identifier | Notes | | --------------- | ------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Google News** | Search terms | We recommend 1 to 3 search terms; each one runs as its own news search. Multi-language search supported (English, French, Spanish, German, Italian, Portuguese, Dutch, Japanese, Korean, Arabic, Chinese). | #### AI assistants The same wizard can also query **Claude, ChatGPT, Gemini, Perplexity, Grok, DeepSeek, and Mistral** about your brand. That's a separate source type with its own analytics: see [AI Visibility](/boundaryai-docs/bringing-in-your-feedback/ai-visibility.md). {% hint style="info" %} **Don't see your source?** Tell your BAI Analytics representative. Additional sources can be wired into the same flow. {% endhint %} *** ### Creating an agent The agent wizard is designed around what you actually want to watch, not the technical configuration. The flow is: describe your goal, BAI Analytics suggests the right platforms and identifiers, you confirm or adjust. #### Step 0: Define Goal When you click **New Agent**, the first screen asks **"What are you looking for?"**. Describe what you want to track, in plain language. Examples: * *"Find all customer feedback and complaints about Hotel Nimbus."* * *"Monitor what people are saying about Acme CRM."* * *"Track brand sentiment for our new restaurant in Barcelona."* BAI Analytics parses the prompt, suggests **which platforms make sense for the goal**, and pre-fills URLs, handles, or search terms where it can. You'll choose the platforms and the exact dates in the next steps, so there's no need to mention them here. {% hint style="success" %} **Tip.** A specific prompt that names the brand, product, or place produces stronger suggestions than a generic one. *"Customer feedback"* gives the AI no anchor; *"Customer feedback on the BAI Analytics dashboard"* lets it suggest concrete handles and queries. {% endhint %} The goal you type here does double duty: it is also what the agent uses to decide whether each collected post or comment is actually about your topic (see [Keeping results relevant](#keeping-results-relevant) below). It stays editable on the Sources step. #### Step 1: Choose Sources The Sources step shows every supported platform, **grouped by how much input you need to give**: * **Runs on your query**: Reddit, X, TikTok, Google News. You provide search terms, hashtags, or handles (or let the AI suggest them with **Suggest with AI**); BAI Analytics handles the rest. * **You provide the links**: Google Reviews, Trustpilot, TripAdvisor, Avis Vérifiés, Glassdoor, Indeed, Google Play, Apple App Store, Amazon, Instagram, Facebook, LinkedIn, YouTube. Paste the page URLs, or tap **Suggest URLs** to let the AI find candidates from your goal. Each platform has a **How to get the URL** helper. * **Ask AI models**: Claude, ChatGPT, Gemini, Perplexity, Grok, DeepSeek, Mistral. Selecting these switches the wizard to the brand-first [AI Visibility](/boundaryai-docs/bringing-in-your-feedback/ai-visibility.md) flow. For each platform you enable, an input lets you paste URLs, handles, or keywords. Inputs become validated chips on Enter, comma, or paste, so you can drop in a list of ten URLs at once and get instant feedback on which ones the platform recognised. Hover a chip to check it, or click a link to verify it opens the right page. A yellow chip means the URL is valid but may not be about your topic (for example, an AI-suggested Glassdoor page for a different company); remove it if it was added by mistake. When the AI suggests URLs or keywords, a short review step asks you to confirm them before continuing. Off-topic entries pull irrelevant data into your analysis, so it is worth a quick check. A small progress counter on each group ("2 of 3 ready") tells you when you have enough sources to proceed. {% hint style="info" %} **One agent, many platforms.** A single agent can watch any combination of platforms simultaneously. There's no need to create one agent per source. {% endhint %} #### Step 2: Set Behavior Once your sources are in, the **Schedule & settings** step covers when the agent runs, how much it collects, and how it's analysed: * **Schedule**: choose **Run Once**, **Scheduled**, or **Ongoing**. For **Scheduled** and **Ongoing**, set **Frequency**, the time in **UTC**, and the day of week or day of month when needed. See [Choosing how often an agent runs](#choosing-how-often-an-agent-runs). * **How far back?**: the look-back window for the first run (**Last 7 / 30 / 90 Days**, **Last 1 Year**, **All Time**, or a custom range). * **Data volume**: the **Scraping Intensity**, *Standard*, *High*, *Extra high*, or *Custom*. See Intensity tiers below. The step shows an estimate of items and credits before you save. * **Analysis**: the **Analysis Language** (sets the source language for sentiment, theming, and translation) and which of the group's monitors will watch the collected data. * **Auto top-up**: optionally add credits automatically so an ongoing agent never pauses. Save the agent and BAI Analytics lands it inside the Feedback Group you created it in, or inside *Individual Surveys* if no group was selected. The agent appears as a card with platform badges, last-run status, and a **Run Now** button. *** ### Intensity tiers Intensity controls how much data a single run pulls. Higher tiers cost more credits but go deeper across all selected platforms. | Tier | Items per platform (typical) | Comments per post | Credits (minimum balance to start) | | -------------- | ---------------------------- | ----------------- | ---------------------------------- | | **Standard** | up to 250 | 10 | 5 | | **High** | up to 2,500 | 20 | 25 | | **Extra high** | up to 5,000 | 50 | 75 | | **Custom** | up to platform max | configurable | calculated from your settings | Each platform has its own cap per tier (high-volume sources like X or Google Reviews go far above the typical figure; page-based platforms like LinkedIn, Facebook, Instagram, or Amazon stay well below it), and *Custom* is bounded by those limits. You don't need to memorise them: the wizard caps your input automatically, and both the pre-run estimate and the credit check use the same per-platform caps, so the run can never demand more credits than the estimate shows. {% hint style="success" %} **Pick Standard for first runs.** If a Standard run already returns the breadth of feedback you need, you have no reason to pay for High. If you see "more available" indicators in the analysis, scale up. {% endhint %} *** ### Per-source parameters Each platform exposes a small number of platform-specific options that the wizard surfaces when relevant: * **Reddit**: subreddits, sort (new / hot / top / relevance / comments), post type (link / self / all), minimum score, max comments per post. * **X (Twitter)**: hashtags, account handles, search terms, max posts. * **Google Reviews / Trustpilot / TripAdvisor / Avis Vérifiés**: list of place / company / listing URLs, max reviews. * **Apple App Store / Google Play**: list of app URLs, max reviews. * **Amazon**: list of product URLs, max reviews. * **Glassdoor / Indeed**: list of company review-page URLs, max reviews. * **Instagram / Facebook / LinkedIn**: list of profile, page, or post URLs. * **YouTube**: video URLs, or channel URLs (up to 10) to pull every video in a timeframe. * **TikTok**: account handles, hashtags, video URLs, or search terms. * **Google News**: search terms; multi-language search across 11 languages. The wizard validates each input against platform-specific rules. For example, Trustpilot URLs must follow `trustpilot./review/`, Glassdoor URLs must carry the company id, and Amazon product URLs are reduced to their `/dp/` form, to prevent typo-driven mismatches. *** ### Keeping results relevant Public platforms are noisy: a brand name that is also an ordinary phrase, a competitor's event in the same city, or a hashtag that spills into unrelated conversations. BAI Analytics filters what an agent brings back so your analysis stays about your topic: * **Every post is checked against your goal.** Each collected item is scored on whether it is actually about the brand, product, place, or event named in your monitoring goal. Content about a same-category but different entity is dropped rather than imported. * **Comments are checked too.** Comments and replies are scored on their own text, with the parent post as context, so emoji-only replies and off-topic threads no longer flood a source. * **Review sites are imported as-is.** Reviews on a page you chose (Google Reviews, Trustpilot, an app listing, and so on) are by definition about that business, so they are never filtered. A comment-heavy source will therefore import noticeably fewer items than a raw count of everything matching the keywords. That is the junk being kept out, not data being lost. #### Marking a comment as not relevant If something slips through, open the source's **Raw Data** tab (the **Feed** view) and click **Not relevant** on the comment. The card dims, shows a **Not relevant** chip, and offers **Undo**. Each mark is recorded together with the search that surfaced the comment, and the BAI Analytics team reviews these cases to improve the agent's filtering. Marking is available on scraped comments only. *** ### Running an agent #### Choosing how often an agent runs When you set up an agent, the **Schedule** section asks **"How often should this run?"** You have three choices: * **Run Once**: collect feedback on demand. * **Scheduled**: run automatically on a schedule. * **Ongoing**: run automatically on a recurring schedule; only new data is scraped each run. For **Scheduled** and **Ongoing** agents, pick a **Frequency** (**Daily**, **Weekly**, **Monthly**, or **Every 3 months**) and the time of day in **UTC**. For **Weekly**, also choose a day of week. For **Monthly** and **Every 3 months**, choose a day of month. The agent card then shows **Next: …**. {% hint style="info" %} **Tracked groups sync to their periods.** When you create an Ongoing agent inside a Feedback Group that tracks feedback over time (Evolution), the frequency pickers are replaced by a single recommended schedule, **Synced to this group's periods**: new posts are fetched every day for weekly periods (every week for monthly or quarterly ones), and each period is completed right after it closes, so no period is left incomplete. The agent card shows a **Synced to periods** badge. Click **Customize schedule** if you'd rather set the frequency yourself; you can switch back to the recommended schedule at any time. {% endhint %} #### First run vs later runs The look-back window you choose when creating an agent applies to the **first run only**. After that, each **Scheduled** or **Ongoing** run collects only the **new data since the previous run**, so you do not pay to re-scrape the same posts. For **Ongoing** agents, a custom look-back is a single **start date** rather than a closed range: the first run reaches back to that date, and every later run just picks up what's new, so no end date exists. The agent card shows an open-ended look-back as *Since {date}*. Only **Run Once** agents need a start **and** end date. Between scheduled runs, **Fetch latest data** on the agent card pulls only new items since the last run without touching the schedule. #### Where a re-run's data goes If you manually re-run an agent that already has a source in the Feedback Group, you're asked **"Where should the new data go?"**: * **Add to existing source**: merge the new items into the same source. * **Overwrite existing source**: replace what's there with this run's results. * **Create a new source**: keep this run as a separate source alongside the old one. Sources feeding an evolution-tracked group skip this choice: they're always append-only, since evolution needs every period's data kept intact. #### Keeping ongoing agents running with Auto top-up The agent wizard includes an **Auto top-up** option: > Automatically add credits so an ongoing agent never pauses. Scheduled and ongoing agents pause if your **Web & Social Media Credits** run out. Turn on **Auto top-up** to add credits automatically when your balance drops below a threshold you set. Auto top-up needs a saved payment method. If a scheduled run cannot secure enough credits, the run card shows **Scheduled run did not start**. The reason explains what blocked the run. Possible messages are: * *Insufficient credits, top up to resume.* * *Auto top-up was attempted but credits are still insufficient.* * *No payment method on file for auto top-up.* * *Auto top-up failed. Please verify your card.* #### Real-time progress Once a run starts, a progress card shows: * A **percentage bar** for the overall run. * The **current platform** being scraped and the items collected so far. * A **per-platform status grid**: checkmarks for completed platforms, spinners for in-progress, an X for any that errored. * Any **error messages** with a platform-specific code. You can keep working in another tab; progress is tracked centrally and the toast follows you. #### Stuck and failed runs A run that stalls is automatically marked as failed so you can start a fresh one. Credits deducted at the start of a failed run are refunded automatically. If a platform's relevance check cannot complete, that platform is marked failed and re-runnable rather than importing a quietly thin result. #### Cancelling Click **Stop Agent** on the agent card during a run to stop it early. Already-scraped data is kept; in-progress platforms are aborted cleanly. #### Retrying If a run fails or partially fails (e.g. one platform errored while two succeeded), the execution history offers a one-click **Retry**. *** ### What scraped data looks like in a Feedback Group Each scraped item (a review, a post, a comment, a news mention) becomes a structured response in the destination Feedback Group, mapped as follows: | Field | Source | Used for | | ---------------------------- | ------------------------------------------ | ------------------------------------------------------------------------- | | **Response** *(Long Answer)* | The main body text (review, post, comment) | Theme detection, sentiment, AI analysis. | | **Source** *(Metadata)* | The platform name (e.g. *Google Reviews*) | Segmentation: filter analyses by platform. | | **URL** *(Metadata)* | Direct link to the original post | Click-through from the analysis to the source. | | **Rating** *(Metadata)* | 1 to 5 star rating (review platforms) | Review-site sentiment cross-check; used in the rating-distribution chart. | | **Likes** *(Metadata)* | Engagement count | X / Twitter only; available as a segmentation dimension. | Because everything ends up as standard answers, scraped data participates fully in **Grouped Themes (previously called Super-Themes)** at the Feedback Group level, can be **filtered by date or sentiment** in the analysis view, and can be **rolled into reports** alongside survey responses and connector tickets. The average star rating shown in the analysis header only appears when the source includes a review platform (Google Reviews, Trustpilot, TripAdvisor, Avis Vérifiés, Glassdoor, Indeed, Google Play, Apple App Store, Amazon); social-only sources carry no star score. Monitoring is not configured in the agent wizard: the group's monitors with **Auto-cover new sources** on watch new agent sources automatically, and the wizard shows which ones will apply. See [Custom Monitoring](/boundaryai-docs/analysing-your-feedback/custom-monitoring.md). #### The Data Source card On the qualitative overview of an agent's source, a *Data Source* card surfaces: * A **platform breakdown** with per-platform item counts. * A **rating-distribution chart** (1 to 5 stars) for review platforms, a quick visual on overall sentiment from review sites. * The **last collected** date. * A **failed-platforms warning banner** if any platform errored on the most recent run, with the list of which ones. * Direct links to **Manage agents** and **Raw data**. *** ### Replying to what you collect Every scraped comment or review carries a **Reply** button wherever it appears in the analysis: the Raw Data tab, the per-platform view, Thematics, Custom Monitoring, and the evidence panels of grouped themes and monitors. Clicking it takes you to the right place to respond, with an AI-drafted reply to start from: * **Suggested reply.** As soon as you open Reply, BAI Analytics drafts a short reply grounded in the review's text and tone (one or two sentences, in the review's language, no boilerplate). Edit it freely; it is labelled as an AI-generated suggestion to review before posting. If a suggestion cannot be generated, you simply write your own. * **X (Twitter)**: continuing opens X's own reply composer with your draft already filled in. * **Reddit, Facebook, Instagram, TikTok, LinkedIn, YouTube, Google Reviews**: continuing opens the exact post or comment in a new tab; copy your reply and paste it there. * **Trustpilot, TripAdvisor, Avis Vérifiés, Glassdoor, Indeed, Google Play, Apple App Store, Amazon**: replies are posted from your business or developer account on the platform, so the modal quotes the review, offers **Copy**, and opens your business console. A secondary link shows the review itself where the platform provides one. * **Google News** and **AI answers** have nothing to reply to, so no button is shown. Your platform credentials never pass through BAI Analytics: you sign in on each platform's own site. *** ### Managing agents Each agent's card shows a running total: the full count of items collected across every run to date, not just the most recent one, so you can see a source's overall size at a glance. Each agent appears as a card on the **My Agents** tab with quick actions: * **Run Now**: start a new execution; the intensity selector and a credit estimate pop up first (**Confirm & Run**). * **Fetch latest data**: pull only new items since the last run, without changing the schedule. * **Configure**: re-open the wizard with the agent's saved configuration. Intensity is auto-detected from the saved limits, so a custom-tuned agent doesn't lose its settings on edit. * **Rename agent**: the new name is applied to the agent, its source, and its analysis labels. * **View Analysis**: jump to the source's analysis page. * **Stop Agent**: stop an in-flight run. * **Save as Template**: see Templates below. * **Delete**: permanently remove the agent. The card updates automatically while a run is live, so you can watch progress without refreshing. #### Execution history Every agent has an **execution history** view with: * One row per run, showing started time, items collected, items imported (after dedup and relevance filtering), status, and a brief error summary if any. * A **per-platform breakdown** for each run: which platforms succeeded, which failed, item counts for each. * A **Retry** button on failed or partially-failed runs that re-uses the same configuration. This is where you go to investigate "why did fewer items come in this week" or "why did Trustpilot stop returning data" without leaving BAI Analytics. *** ### Templates If you build an agent configuration that works well (say, a multi-platform brand-listening setup) you can save it with **Save as Template** on the agent card. Templates: * Become **clonable** within your organisation from the **Templates** tab, so a teammate can spin up a similar agent in seconds with **Use Template**. * Are version-independent. Editing the original agent does not retroactively change clones. * Are useful for **standard playbooks** (event-monitoring, product-launch listening, competitor tracking) you expect to re-run. *** ### Platform status A **Status** button on the Agents page opens the **Platform Status** modal: a real-time health check for every supported platform, based on runs in the last 48 hours (Operational / Degraded / Down, with the success rate). Use it to diagnose whether a poor run was your configuration or a temporary platform-side issue. A **Refresh** button forces a fresh check. *** ### How agents differ from connectors and uploads | Feature | Use it for | Auth | Cadence | | ---------------------- | ----------------------------------------------------------- | --------------------- | -------------------------------- | | **Agents (this page)** | Public data: reviews, social posts, news | None (web automation) | On demand, Scheduled, or Ongoing | | **Connectors** | Data inside tools you own: support, dev, project-management | Secure sign-in | Continuous (30-min) | | **Uploads** | Historical or third-party exports | None | One-shot | | **Surveys** | Direct outreach to known respondents | Optional passcode | On respondent submission | You can mix all four inside a single Feedback Group, and BAI Analytics' analytics treat them uniformly. *** ### Best practices * **Start with a focused goal.** *"Brand sentiment for X across major review sites and Twitter"* yields a tighter, more useful agent than *"Customer feedback."* The goal also drives relevance filtering, so name the exact brand, product, or event. * **Pick the right intensity for the cadence.** A weekly Standard run on a small brand is usually enough; a daily Standard run on a large brand may miss data. Go High instead. * **Use a dedicated Feedback Group per topic.** Don't mix brand listening with employee feedback in the same group; the Grouped Themes will be muddled. Glassdoor and Indeed agents belong in their own group. * **Validate sources after the first run.** The Platform Status modal and the execution-history view tell you which platforms are returning data. If one platform is consistently empty, double-check the URL or handle. * **Mark what slipped through.** A few **Not relevant** marks on off-topic comments help the team tune the agent's filters. * **Treat *****Custom***** intensity as the exception, not the default.** The three preset tiers cover most use cases; *Custom* is for power users who know exactly which platform-specific limit they want to push. * **Save proven setups as templates.** A team running brand listening across multiple products will build the same configuration repeatedly. Templates remove the re-work. * **Refund-aware retries.** Failed runs refund credits, so re-running a failed agent is safe. You won't be double-charged.