# FAQ > Skald docs: https://skaldanalytics.com/docs/faq/ . Every guide: https://skaldanalytics.com/llms.txt . Skald is source-available (Elastic License 2.0). Short answers to the questions people ask most about Skald's figures, privacy and data. ## Why do my numbers differ from Google Analytics? They count different things, in different ways. - Google Analytics counts only browsers that run its script and accept it. Ad blockers, privacy browsers and refused consent banners hide many visits from it. Skald reads the server's own log, so it sees those visits. - Google Analytics cannot see AI assistants, search engines or bots. Skald counts them, but in their own groups, never as people. - Skald is careful about who is a person: when in doubt, a request is not a person. Its people figure is an upper bound. - On sites that change pages without loading a new one, the log sees fewer page views than a script does. Skald says so where it matters. - Days may be cut at different times if the two tools use different time zones. ## Why does a day say withheld? Every figure is counted twice, by two separate roads. If the two counts do not agree, the day is withheld rather than shown wrong. It is never shown as zero. The server admin can see why on Site > Data quality. ## Why is a figure blank and not zero? Blank means not known. Zero means Skald counted and found nothing. A figure that was not recorded, for example a breakdown from before Skald began collecting it, stays blank. What was not recorded is never guessed. ## Why do my countries say Not recorded? Skald takes a visit's country from your server's log. Some logs carry it (Cloudflare's country, or nginx's own `cc=` field). Others do not, for example Caddy's log or nginx's stock log. For those, whoever runs your Skald server can set a GeoIP country file (`logs.country_db`, or `SKALD_COUNTRY_DB` in Docker; DB-IP's free file works). Skald then looks up each visit's address in it. Without a file, the country stays Not recorded. It is never guessed. See [Docker: first install](https://skaldanalytics.com/docs/docker-first-install.md). ## Is AI traffic counted as people? No. AI crawlers and fetches are machines, and show under AI & bots (Assistants and Crawlers) in their own units. A person who clicks a link in an AI answer is a person, and the AI assistant is shown as their source on Traffic > Sources, and in full on AI & bots > Assistants. ## What does "likely from ChatGPT" mean? It is an estimate, not a count. A person who clicks a link in an AI answer often arrives with no referrer, so the visit shows as Direct. But when they asked, the assistant fetched your page from your server. When a direct visit with no campaign tag lands on a page an assistant fetched in the same minute or up to 15 minutes before, Skald says the visit was *likely* sent by that assistant. It shows these on AI & bots > Assistants, on Leads & sales > Sources (with either Count from) and in a visitor's details, always marked "likely" with an (i) that says how. They are never added to the AI assistant source or to any other figure. See [Leads and conversions](https://skaldanalytics.com/docs/leads-and-conversions.md). ## What did the AI read before it sent this visitor? Open a visit on Traffic > Visitors. If the visit followed a link in an AI answer (ChatGPT, Perplexity, Claude, Gemini, Copilot, Mistral, Meta AI), or is likely from one, **What the AI read before this visit** lists the pages that assistant's user fetcher fetched in the 15 minutes before the visit: "Fetched by ChatGPT in the 15 minutes before", nearest first, at most 10 pages, with the page the visit landed on marked. It is context, not a count of people: a fetch can be for someone else. Training crawlers (GPTBot, ClaudeBot, ...) are never listed: nobody was waiting on them. On AI & bots > Assistants, **What the AI read before it sent a visitor** shows, per assistant, the visits from its answers, how many came after it fetched pages, and how many landed on a page it had just fetched (fetch, then a visit to the same page). Likely visits are shown apart: they always land on a page the assistant fetched, because that is how they are inferred. Assistants with no known user fetcher (You.com, Phind, Poe) are not in that table. Days before this feature read "not recorded". The window is the same setting as for "likely" (`attribution.ai_window_minutes`). ## What are the AI alerts? Three alerts watch how AI assistants treat your site. They show on Site > Alerts (press **AI** to see only them) and in **AI reading changes** on AI & bots > Assistants: - **AI started reading**: an assistant fetched a page yesterday that it had not fetched in the 28 days before. - **AI stopped reading**: an assistant that fetched a page in at least 3 of the 4 weeks before has not fetched it for 7 days. - **AI sending people**: the visits likely sent by an assistant jumped (3 times its 28-day average, and at least 10) or vanished. With fewer visits Skald says "too few to judge" and does not alert. An assistant means its company: GPTBot, OAI-SearchBot and ChatGPT-User are all OpenAI. Each alert names the assistant, the page, the days and the numbers. They use whole days only, and only figures Skald has already counted twice. If a day in the window is withheld or missing, the alert waits; it never fires on a day it cannot trust. The server admin can change the thresholds or switch each one off (`alerts.ai_*` in the server's config). If alert mail is on, the weekly mail lists the week's changes, and an outgoing webhook can receive each one as `ai.reading_changed` ([Webhooks](https://skaldanalytics.com/docs/webhooks.md)). Four more alerts watch AI citations and crawlers: - **First AI citation**: the first visit ever that followed a link in an assistant's answer (ChatGPT, Claude, Perplexity, ...). The alert gives the time, the page it landed on and that day's visits from the assistant. - **A page's first AI citation**: the first visit ever that landed on a page from an AI answer. More than 5 pages on one day make one alert that lists them. - **Visits from AI answers**: yesterday's visits from AI answers compared with the usual for that weekday. The usual is the median of the same weekday over the 6 weeks before (at least 4 of them counted). The alert fires when the day has at least 2 times the usual and at least 10 visits, or at most half of a usual of at least 10. With fewer visits Skald says "too few to judge" and does not alert. The alert lists each week's figure, so you can check it. - **Crawler gone silent**: an AI or search crawler that came on at least 7 of 28 days has made no request for 7 days. Check robots.txt and any firewall or CDN rule. These count visits (a person who clicked a link) or requests (a crawler), never both in one figure. The first-citation alerts start once Skald has 28 days of history, so a new install does not call every citation a first. The morning run checks each alert's figures again against the stored days, with separate code. The server admin sets these in the server's config too (`alerts.first_citation`, `alerts.cited_weeks`, `alerts.silent_days`, ...). To get any alert in Slack or another tool, add an outgoing webhook with the event `alert.opened` ([Webhooks](https://skaldanalytics.com/docs/webhooks.md)). ## What do AI assistants and crawlers fetch? **What they fetch** on AI & bots > Assistants (fetches for a question) and Crawlers (fetches for training) splits every AI fetch by what its address names: a page, your llms.txt (`/llms.txt` or `/llms-full.txt`), a Markdown file (`.md`), another text file, robots.txt, a sitemap, or another file (images, scripts, feeds, PDFs). A second table gives the same split per assistant or crawler; each day keeps the 20 with the most fetches and sums the rest in "(all others)". The address decides, not the answer: a fetch of an llms.txt you do not have still counts as llms.txt. Days counted before version 2.10.0 have no split, and a range that holds one shows none. ## Can Skald see Google's AI Mode? Only when Google says so. A click from Google whose address carries `udm=50` (Google's AI Mode) is still a search engine visit, but its referring site reads **google.com (AI Mode)** on Traffic > Sources, so you can see it apart. Most clicks from AI Mode and AI Overviews arrive as a plain `google.com`, the same as any search click; Skald never guesses which those were. ## Can Skald write an llms.txt for my site? It suggests one; it never writes it to your website. Open AI & bots > llms.txt and see **Suggested llms.txt**: links to the pages people and AI assistants read most over the last 90 days, grouped by section, in the llmstxt.org format. Write the one-line summary yourself, check the links, then copy or download it and publish it at `/llms.txt`. See [Settings and retention](https://skaldanalytics.com/docs/settings-and-retention.md). ## Why are pages opened inside my Next.js site not counted? A click on a link inside a Next.js site loads no new page from your server: the browser fetches only the next page's data. Next.js fetches the same data for links on screen that nobody clicked (a prefetch). Unless the server's log records the header Next.js sends with a prefetch, Skald cannot tell a click from a prefetch, so it counts neither. Only the first page of each visit shows, and a visit from a tab left open may count as no visit at all. One line in the web server's log format fixes it: see [Troubleshooting connections](https://skaldanalytics.com/docs/troubleshooting-connections.md), "Next.js sites". Overview's checklist shows **Pages opened inside the site are not counted yet** while it is missing. ## Why is my own visit not counted? Your team's addresses are probably listed as internal. Visits from them go to the "Our own" group, kept apart from people, so your own testing does not inflate the figures. See [Internal traffic](https://skaldanalytics.com/docs/internal-traffic.md) and Internal traffic settings. ## Does Skald use cookies? Not to count visits. Skald reads the server's log, which needs nothing in the browser. The optional script tag sets no cookie by default; a cookie is used only if your site turns on anonymous visitor IDs shared across subdomains (on the tag, or in Skald's settings). Signing in to the dashboard uses one cookie, for your own session. See [Privacy](https://skaldanalytics.com/docs/privacy.md). ## Does Skald count visitors who decline or ignore the cookie banner? Yes. Skald counts from your web server's own access log, and the server writes a line for every request, whatever the visitor chose on the banner. Skald sets no cookie and stores nothing on the visitor's device to count them. The optional script tag is the one part that follows the banner. It can wait for consent (`data-consent="required"`, then your banner calls `skald('consent', true)`), it does nothing when Do Not Track or Global Privacy Control is on, and it keeps no visitor ID unless you turn IDs on for the site. The log holds IP addresses, which are personal data in places like the EU. You choose whether Skald keeps them in full, shortened (the last part removed) or not at all, and for how long (Data retention). The same page writes the wording for your privacy notice from those settings. See [Privacy](https://skaldanalytics.com/docs/privacy.md). ## How fresh is today? Today's figures are recounted about every two minutes. The foot of every page says when they were last updated and when the next update is due. If your server allows it, **Refresh** in the top bar recounts at once. Yesterday becomes a whole, checked day after the morning's run. ## Why does a whole day appear only the next morning? The daily run checks each finished day by both counts before it is stored. Until then, today's figures are a live count, and can change. ## Where are my leads? On Leads & sales > Leads: every lead of the period, one row each, with its name and email for owners and admins (when your site keeps them), what they did, its stage now and where it came from. Summary has how many against the period before, and Sources where they came from; both count from your lead database by default; switch **Count from** to **Tools** for bookings, forms and payments from your tools. Leads and demos booked also show beside the visitors on Overview > Summary, Traffic > Sources, Pages and Places & devices, and Overview > Insights. Spam, duplicate and test leads are listed on Leads with **More > Show left out**. Owners and admins can find one lead, with its email and the visits around it, on Leads & sales > Find a person or with the search at the top, when your site keeps names and emails. See [Leads and conversions](https://skaldanalytics.com/docs/leads-and-conversions.md). ## How do I get data out? - **Export what you see** above any table: its matching rows, in their order, with the columns shown, as a CSV. - **Export** in the top bar: every table on the tab as CSV files in a ZIP, the tab as JSON, a printable report, or the report builder for several tabs at once. - **API**: every figure as JSON, for a script. Tokens are made in **Settings > Team & access > Connections**. See [API](https://skaldanalytics.com/docs/api.md) and [Connections](https://skaldanalytics.com/docs/connections.md). - **MCP**: let an AI assistant read your figures. See [MCP](https://skaldanalytics.com/docs/mcp.md). ## Who can see visitor addresses? Only people signed in to your organisation: Site > Requests (the last 6 hours), today's AI activity (on AI & bots > Assistants, when the period is Today), and the internal-address list (admins). When your site keeps IP addresses, owners and admins also see each kept visit's address in the person search; every search goes to the audit log. Stored history never holds an address. API tokens see addresses only with **Reading visitor addresses**. See [Privacy](https://skaldanalytics.com/docs/privacy.md). ## How do I delete someone's data on request? Send the address or the email address to the person who runs your Skald server. They run Skald's forget command, which deletes the visits, conversions, name and email kept for it. See [Privacy](https://skaldanalytics.com/docs/privacy.md). ## Why are my filters greyed out? The log keeps each breakdown on its own, so the counted tabs apply one value per breakdown and cannot cross them (for example, Canada AND Chrome). Rules over single visits (several values, OR, groups, pages read, time on site, first came from) apply in full on Traffic > Visitors and Leads & sales > Funnels, and a crawler rule on Site > Requests. A greyed-out chip is one the tab cannot apply; a line at the top of the tab says which, and links to where it does apply. ## Can I see times in my own time zone? Yes. Times show in the site's own zone unless you choose otherwise. Select the round **You** button at the foot of the sidebar and choose **Time zone**: the site's, your own, or UTC. Days are still counted midnight to midnight in the site's own zone. ## How do I invite a colleague? Admins and owners go to **Settings > Team & access > Team** and use **Invite someone**. See [Sign-in and team](https://skaldanalytics.com/docs/sign-in-and-team.md). ## My site is behind Cloudflare: will cached pages be missed? Only if Skald reads nothing but your server's log. A page Cloudflare serves from its cache never reaches your server, so your server's log does not see that visit. Two fixes, set up by whoever runs your Skald server: Cloudflare's own request log (Logpush), if your Cloudflare plan offers it, or the small Skald Worker on any plan. The Worker sends Skald one line per request after the page has gone out, so it never slows the page. When Skald reads both the Worker and your server's log, it takes only the cached pages from the Worker, so no visit is counted twice. ## What does "A newer Skald is available" mean? A newer release of Skald is out than the one your server runs. Only owners and admins see this line at the top of the dashboard, and the daily and weekly mail show it too. **What's new** opens the release notes. **×** hides the line in your browser until the next release. A red line is a security release: ask whoever runs your Skald server to upgrade soon. Nothing changes until they do (or until they switch on automatic upgrades). To check, your server asks GitHub once a day which release is the latest; it sends only Skald's version, never your figures or visitors. Your server admin can turn this off (`updates.check: false`). See [Upgrading](https://github.com/spcaeo/skald/blob/main/docs/UPGRADING.md). ## How is Skald upgraded? Whoever runs your Skald server upgrades it with one command. It checks the download, copies every database first, refuses to jump to a new major version or to run during the morning run, and puts the old version back if anything fails. Your figures and settings stay. Automatic upgrades for fixes (or fixes and features) can be switched on; they are off unless your admin turns them on. With Docker, the admin pulls the new image. See [Upgrading](https://github.com/spcaeo/skald/blob/main/docs/UPGRADING.md).