Know what AI and people say about your brand

It answers a question no traditional tool does: when someone asks an AI assistant about your brand instead of searching, what do they hear? We ask twelve assistants (ChatGPT, Claude, Grok, Gemini, Perplexity, Mistral, DeepSeek, Kimi, Microsoft Copilot, Brave Search, Google AI Overviews and Google AI Mode) the questions real buyers ask: best in your category, is this brand good, what do people complain about. Then we report whether you were recommended, the exact words used, which competitors were named instead, which websites the assistants cited, and how much of each answer came purely from the model's memory. It ends with concrete actions drawn from your own results.

Yes. Open the finished check and choose Ask Polly at the bottom right. Polly answers from the questions, saved answers, sources and to-do list in that check. It does not run a new check. Review the linked evidence before acting on an answer.

Yes. In the Brand hub, Your review holds Approved brand facts and your saved writing voice. Drafts and rewrite advice are instructed to follow them. You can fill in a missing fact in a draft and remember it with a public source for later use. This is writing assistance, not a check of every sentence: review facts, claims and wording before publishing.

Yes. Open Your website in the Brand hub and switch on Site watch. Once a day it reads up to 20 pages of the website in your check and alerts you to important changes, such as a broken page or access being blocked. It uses no credit and asks no assistant. It follows robots.txt, can be stopped, and pauses 90 days after the brand's latest check. It is a daily check, so a brief outage between looks may go unseen.

A growing share of buying decisions starts with an AI answer rather than a search results page, and those answers are invisible to you unless you go asking. An assistant can be recommending your competitor, describing you with years-old information, or leaving you out entirely, and no analytics tool will show it. Knowing what the AIs say, and which sources they trust, is becoming as basic as knowing your Google ranking was. We also check DeepSeek and Kimi, the assistants used by hundreds of millions of people in China and beyond, which most Western tools ignore entirely.

Almost certainly not word for word, and that is expected rather than a fault. These models are non-deterministic: ask the same question twice and the wording, the order of brands, even which competitors appear can change. Replies also shift with the account asking, the country it is asked from, and the day. That is precisely why a check asks 12 buyer questions across twelve assistants rather than spot-checking one: a single answer is noise, the pattern across more than a hundred is the signal. The saved answer text and its date let you inspect the evidence; older checks may retain an excerpt, and each check is a measurement of that moment, not a permanent verdict. Two more things worth knowing: what a check receives can differ from what the assistant's own app shows a signed-in person, in wording and occasionally in the brands named (DeepSeek is asked in its standard non-thinking mode, the one its app uses unless a user switches DeepThink on); the check shows observed shares for its question set without claiming a statistical margin of error, while a tracked brand's page shows how consistent each answer has been across its recent checks.

Live systems have bad moments: an AI provider can be briefly overloaded or rate-limited while your check runs. Usually that costs one or two questions out of the twelve we ask that assistant, not the whole assistant. Either way the affected questions say "no answer" honestly, and your check tells you exactly how many an assistant answered, and we automatically re-ask just the missing questions 15 minutes, 3 hours and 9 hours later, at no extra cost and with nothing for you to do - your first result is usable within about an hour, and we email you when the check is complete. If an assistant is still unreachable after that, a "Re-ask missing" button on your check page starts the whole cycle again, free, and only for the questions still without an answer. What we never do is invent an answer to fill a gap: if an assistant genuinely returns nothing, your check shows that, because it is exactly what a buyer asking that assistant would experience.

Your brand checked every week across 12 AI assistants, with what they read, what they get wrong, and what to publish to fix it. Most trackers ask three or four assistants a long list of questions every day and chart how often you are mentioned. PollyX goes deep instead: each check asks 12 buyer questions to all twelve assistants and reads the pages behind every answer, so you see why you are or are not recommended, which facts the assistants get wrong, and what to publish to change it. Every check asks every question live, at the moment it runs, and no answer is reused from a cache or an earlier check. Turn tracking on from any check (weekly, every two weeks or monthly); every check on a brand gathers on its Brand page.

The sentiment report measures what PEOPLE say: news, videos, reviews, forums and social posts, read and scored. The AI Visibility check measures what MACHINES say when asked. They complement each other: public opinion feeds the AI's training data today, and the AI's answers shape public opinion tomorrow. Both cost 1 credit.

Yes, as paid add-ons, and you see the added price before you pay. An extra question costs €2.19 a check, asked of all the assistants, or €0.50 asked of one you choose (up to 12 extra questions). A city costs €8.09 a check: four questions about that city (up to 5 cities). Daily tracking costs €59 a month per brand: five of your questions asked every morning of ChatGPT, Perplexity and Google's AI Overviews, with an alert the same day when something moves. On a tracked brand, extra questions and cities are billed monthly at the same per-check prices. When a plan check or a credit pays for the check, its add-ons are a card payment of their own, which starts at €1.99 (a month, on a tracked brand). If a check fails, its add-ons are refunded with it.

Open Evidence behind the findings in a new check. It shows exact claim excerpts from passages across the saved answers, the assistants behind them, available source links and page dates. Repeated wording is counted separately from assistant variety. A cited page is not proof that a claim is true, and limited support is labelled. The same evidence summary appears in the PDF, PowerPoint and Markdown exports.

It flags possible conflicts when matching statements about the same question have different values or negation. For a few unresolved price, policy or product claims, it can review up to 3 already-discovered pages on the supplied official website. You see a related passage, an unavailable page or no match. Review dates and conditions before deciding what is correct; PollyX does not assign a truth verdict.

Yes, when you also have a compatible consumer sentiment report. The reports must match the brand, market and language, cover consumers, and be within the preceding 30 days at the time of the AI check. Linked examples use exact excerpts about the same brand. Opposing views require the same proposition and scope with explicit negation; different surveys or ratings do not qualify. Concerns may be absent from the sampled passages without being absent from all AI answers. Older comparisons without context-checked excerpts hide their theme interpretations. Missing coverage does not prove a claim false. The consumer report uses its own check or credit, and the two products' scores are not compared as if they were the same measure.

A lot: platform-by-platform sentiment, positive and negative themes with real quotes, audience segments (how different communities feel about you - and they often disagree), purchase-intent signals (people saying they're about to buy, switch or leave, in their own words), emotional tone, share of voice against your competitors, a 26-week timeline counting saved news and video records, who's talking (account types, geographies, languages - each labelled by how it was measured), crisis alerts, trending topics, and the improvement roadmap. Counts are calculated from saved records; scores and distributions are AI assessments. Source links support checking the findings, but not every narrative claim has a verified source mapping.

It's a single measure of overall public sentiment, distilled from hundreds of real public sources - global news, YouTube videos and comments, and live web search across review sites (Trustpilot, Google, Yelp), Reddit, forums, and other public platforms. An AI analyst weighs the balance of positive versus negative opinion, considering the strength and recurrence of expressed opinions, recency and source credibility, then condenses it into one number: 7 to 10 is positive, 5 to below 7 mixed/neutral, and below 5 negative. The score is PollyX's own calculation and is not a metric published or endorsed by any of the platforms analysed. It is an AI assessment, not a survey statistic. Reports are scored independently; different evidence and model judgment can change the number without proving a real shift in public opinion.

Yes. You can generate a report on any public entity: your own brand, a competitor, a product, a public figure, or even a topic. You can also add up to 5 competitors in a single report for direct comparison.

By default, PollyX gives you the comprehensive picture - it analyses both consumer/customer sentiment (Trustpilot, Google, Reddit, news, social) and employee/workplace sentiment (Glassdoor, employee reviews), and labels employee perspectives clearly so you can tell them apart. If you only care about customer perception, choose 'Consumers only' when creating the report and we'll exclude all employee and workplace opinions.

Yes. Choose 'Campaign' when creating a report and give us the hashtag - we read the tag's own feeds on X, TikTok and Instagram plus news, YouTube and Reddit reaction, and the report becomes a campaign-reception analysis: how it landed, participation, standout moments with links, and what misfired. Same price as any report, 1 credit.

Yes - it's one of the most valuable parts of each report. Every action in the roadmap is grounded in the actual sentiment data we found (not generic advice), includes an impact/effort rating, and comes with specific step-by-step instructions on how to implement it.

Yes. Every completed report has a built-in chat: ask anything about the findings ('what's driving the negative themes?', 'summarise this for my CEO') and it answers from your report's own data, with quotes. Included with the report at no extra cost.

Most reports are ready in about 15 minutes. The exact time depends on how much data is available for your brand and how many competitors you requested. You'll see a progress screen while it runs, the page opens the report automatically when it's done, and we email you the PDF, so you can close the page in the meantime.

Yes. Choose 'Product' when creating the report and name the model - 'Dyson Airwrap', not 'Dyson'. The evidence then comes from where people actually discuss products: YouTube reviews and their comments, Reddit threads, TikTok, review sites and Amazon ratings, with the score split six ways by aspect (product quality, price and value, delivery, support, ease of use, trust). News coverage is usually thin for a single product and the report says so instead of padding it. Same price as any report, 1 credit. It reads opinions, not prices: for a fair market value use a price guide.

Yes. On the new-report form you can upload a CSV of your own customer feedback: reviews, support tickets or survey answers, up to 200 rows (a text column, plus source, date and rating if you have them). The report reads it as clearly labelled first-party evidence next to the public sources, says where a theme also shows up in your own data, and marks anything it takes from your file as customer-provided. Your file never changes the public sentiment score and is never listed as a public source; it is used for that one report, at no extra cost.

Evidence behind the findings shows the source records linked to each theme, host variety, known dates and known copies of the same story grouped together. Few linked sources mean limited evidence; different hosts do not prove independence. When searches repeat stories or miss the subject, new reports can redirect the remaining search allowance toward missing themes, source types or markets within the requested scope. The total allowance stays fixed, and searches never aim for a particular positive or negative balance.

PollyX searches millions of sources, including global news from 100+ countries; X, TikTok, Instagram, YouTube, Reddit, LinkedIn, Threads, Bluesky, Telegram and Twitch; Trustpilot, Google Reviews, Amazon, the App Store, Google Play, Tripadvisor, Booking.com, Airbnb, Yelp, Deliveroo, G2, Capterra, eBay, Etsy and Steam; Hacker News, Substack, Medium, GitHub, Stack Overflow and Product Hunt; Glassdoor, Indeed and kununu; podcasts; and local platforms such as Bilibili and Zhihu in China, PTT in Taiwan, Kaskus in Indonesia, Reclame Aqui and Mercado Libre in Brazil and Zomato in India. Which of them run depends on the brand's industry and market. In more detail - News: the GDELT Project (global news articles from 100+ countries). Video: the YouTube Data API (video reviews, complaints, and thousands of real comments). AI-powered live web search, which scans Trustpilot, Reddit, Glassdoor, forums and more in real time. Social: public posts from X, TikTok and Instagram (fetched directly, filtered for ads and spam), Threads, Bluesky and Mastodon, Telegram public channels, and the replies and comments under the posts we find. Forums and communities: Reddit, Hacker News, local forums such as boards.ie, Substack newsletters and their comments, Medium and Tumblr. Developer communities, for software brands: GitHub issues, Stack Overflow questions and Product Hunt launch comments. Review, app-store and complaint platforms, read in aggregate and never quoted: Trustpilot, Google Reviews, Amazon, eBay and Google Shopping product ratings, the Apple App Store, Google Play, the Chrome Web Store and the Shopify App Store, Steam reviews for video games, Etsy shop reviews for makers, G2, Capterra, TrustRadius and Gartner Peer Insights for business software, BBB, PissedConsumer and Sitejabber. For hotel and travel brands, read the same way: Tripadvisor, Booking.com, Expedia and Hotels.com guest reviews, Airbnb listings the brand runs, and Yelp for premises. For restaurant brands: Deliveroo and Zomato outlet ratings. Regional platforms, for reports on their markets: Reclame Aqui complaints in Brazil and Mercado Libre product ratings across Latin America, read in aggregate, and Bilibili videos and Zhihu answers in Chinese-language markets, PTT threads in Taiwan and Kaskus threads in Indonesia, described rather than quoted. Employer reviews, unless you ask for consumer sentiment only: Glassdoor and Indeed, plus kununu in Germany, Austria and Switzerland. Investor chatter for listed companies: Stocktwits, described rather than quoted. Podcasts: episode titles and show notes from Apple's podcast index, plus transcripts of the episodes whose title or show names your brand - the publisher's own transcript where the feed carries one, otherwise a machine transcript of up to 25 minutes, which can misrecognise names and numbers and is attributed to the show, never to a named speaker; your own shows are excluded. Attention signals: Google Trends search interest and Wikipedia article activity. Through those indexes the search reaches millions of sources - the same web every monitoring tool sees. Most of it never mentions your brand, and of what does, spam, ads, duplicates and the brand's own marketing are removed before scoring. What survives is what your report actually reads and cites: every source that talks about the brand, after irrelevant, duplicated and spammy items are removed, every one listed and clickable, and every report shows its own exact count on the page.

You're only charged for reports we can actually deliver. If we can't find enough public information to build a meaningful report - which can happen with very new, very small, or private businesses - the report won't generate and your payment for it goes straight back to your original payment method: the full price if you paid by card, or the price you paid for that credit if it came from a pack or a plan. The refund is the payment itself, never a replacement credit, and it usually shows within 5 to 10 business days. When there's some data but it's limited, we still generate the report and mark it clearly as lower-confidence.

The website, every report and check, the PDF and slides, emails and support are in English. An AI Visibility check can ask the assistants in English or in one of 32 other languages, such as German, French, Spanish, Japanese, Chinese, Korean or Russian: you write and read the questions in English, we ask them the way buyers in that market would type them, and the findings come back in English with quotes in the answers' own words. Reports read sources in many languages and quote them in the original with a short English note. Thai, Serbian and Arabic are asked in English for now.

Yes. From any completed check, choose 'Track this brand' and pick weekly, every 2 weeks or monthly. Your brand's questions are saved on the first check and re-run word for word on every later one, so a change in the score is a change in the answers, never in the questions, and you're alerted in-app (and on Slack, Teams or a webhook) when your score moves by 8 points or more, or when you lose first place on a question that decides a purchase (a best-in-category or comparison question). Because each question is asked once per assistant, one check is one sample; across your recent checks the page shows consistency - how often each question led, and how often each assistant named you first on the category question - which is the pattern that matters, not any single week. For a reading every morning, daily tracking (€59 a month per brand) asks up to five of your questions of ChatGPT, Perplexity and Google's AI Overviews once a day and tells you the same morning when you lose first place.

Yes, two ways. Monitoring re-runs your full report on a schedule you pick (weekly, every 2 weeks, or monthly - one plan check or 1 credit per run) and charts your score over time. On top of that, every monitored brand gets a free daily crisis radar: we count the last 24 hours of news and community activity, and if a channel spikes we alert you in-app and on your Slack, Teams or webhook - with a one-line explanation of what's likely driving it.

No. AI Visibility checks and reports are pure pay-as-you-go: €20.99 each, VAT included, only when you need one, no commitment. Monthly plans are coming for teams that want always-on monitoring - Lite at €41 a month for 2 checks, Starter at €102 for 5, Growth at €142 for 7 and Pro at €304 for 15, with extra checks at €20.49 - but they will never be required to buy a check or a report.

One credit buys either product: an AI Visibility check or a full Brand Sentiment Report. Same price, your choice each time. Buy one at €20.99, or save with a pack: 5 credits for €102.00 or 10 credits for €204.00. Credits never expire.

Yes. Prices include VAT: the price shown is the total you pay. Stripe, through its Link service, acts as merchant of record for the sale: it works out the tax included in the price for your country, collects and pays it, and your receipt from Link shows it. Where the checkout lets a business enter a VAT number, the tax is applied accordingly.

Reports and AI Visibility checks are generated on demand, so all sales are final once yours has been delivered - when you order, you tick a box asking us to start straight away and confirming that the 14-day EU withdrawal right for that item ends once generation begins. You are never charged for a failure, and you never have to ask. If we can't find enough public data on your subject to build a report at all, you are refunded immediately and in full, back to your original payment method: the full price if you paid by card, or the price you paid for that credit if it came from a pack or the API plan. No credit stays on your account, and a credit that was granted or earned rather than bought is simply not re-issued. If one of a monthly plan's checks paid for it, that check goes back to your allowance, so the report doesn't count. If a report hits a technical problem instead, you get one free re-run at no cost, and if that re-run fails too, or you simply don't retry, the charge is refunded automatically to your original payment method (or back to your plan's allowance if a plan check paid). In a Full Picture the report comes first: if it can't be made, the AI Visibility check hasn't started yet, and you choose whether to start it or have it refunded too. A card order that can't even start on our side is refunded straight away. Every refund shows up as a notification in your dashboard and by email, and reaches your bank or card within 5 to 10 business days. Credit packs and monthly plans are different, because nothing is used up when you buy them: you can withdraw within 14 days of buying and we refund the credits you haven't used, or, for a plan, the month's price less the checks or credits already used. After 14 days a pack is non-refundable, but credits never expire. Full details are in our Refund Policy.

A refund and your credit balance always agree. When a report can't be built, the money goes back and the credit it used simply stays spent - nothing is re-issued. If we ever refund a pack or a plan payment for another reason, the credits that money bought are removed from your balance at the same time, and you'll see a notification in your dashboard saying exactly what was refunded.

Yes. The Full Picture is a Brand Sentiment Report plus an AI Visibility check on the same brand for €39.99 - one order, €1.99 less than buying them separately. Tick "Add the AI Visibility check" when you create a report. If your account holds credits it uses 2 of them instead. The report comes first, and the check starts as soon as the report is ready. If the check fails, it is refunded and the report stays yours; if the report can't be made, it is refunded and you choose whether to start the check or have it refunded too (we refund it if you don't choose within 14 days).

Every AI Visibility check exports as PDF, slides, CSV or Markdown, with a read-only share link and a badge. Every report exports with one click as PDF, PowerPoint slides, Excel, CSV, or Markdown (which imports straight into Notion, Obsidian or Confluence). You can also share a report's read-only public link, email its PDF to up to 3 colleagues, or post a summary card to your Slack channel. For automation, PollyX has a self-serve REST API and an MCP server - create an API key in your dashboard and your own tools (or your AI assistant) can order and read checks and reports directly. See /docs/api.

Yes. In Settings you can set a company name and upload a logo, and the PDFs you export, of AI Visibility checks and reports alike, carry your brand instead of PollyX's. It's built for agencies, consultants, and in-house teams who buy PollyX checks and reports and hand them to clients or leadership under their own name. Every account has it: fill in the white-label fields under Settings and it applies automatically to every PDF you export - leave them blank to keep the default PollyX branding.

Only if you choose to. Every AI Visibility check and report is private to your account; you can share a read-only link and switch it off at any time. You can also invite a colleague or client to one brand and market. They must sign in with the invited, confirmed email address; they see the latest shared results without private answers, downloads or editing controls. You can remove access. PollyX's public brand pages show short, dated previews that PollyX makes itself, never your reports, and nothing there is sold as a copy: a report you order is always made fresh, with today's data.

Yes. Open the AI Visibility check or report and use Delete at the bottom of the page. It leaves your dashboard straight away and its share link and downloads stop working, but it waits in Recently deleted, at the bottom of your AI checks or Sentiment page, for 30 days: Restore brings it back exactly as it was, and Delete forever erases it at once. After 30 days it is erased automatically. Copies other customers already bought from the catalog stay theirs, and deleting doesn't return the credit or payment it used. A report that is still being generated, or whose refund is still being settled, can be deleted once that has finished, and a report refunded to you stays in your list for 30 days after the refund. Deleting your account in Settings erases everything at once.