Skip to content

For AI agents: this page as Markdown; every page and guide as Markdown is listed in llms.txt.

New storyWe built Skald to count visitors. It saw an AI-key thief at work.
SkaldEvery visit, truly told.

Insights · from reactive to proactive

Twelve questions a site owner asks, answered before they turn into a crisis

A client asks why leads fell, and the digging starts days after the damage. Proactive website analytics flips that order: your own server log is read every day, and what changed, what to fix and whether a fix worked come to you first, with their proof.

Reactive analytics waits for someone to notice. Proactive, as used here, means each worry has a place that already holds the answer: a figure, the days it covers, and the next step. Below are the twelve questions and the places in Skald that answer them, all read from the log; the method behind the log is in how server log analytics works.

By · Updated · 10 min read

one week of looking after a site · sample
flip the week
  • Notice the dropa client asks why leads fell01:00 summary: 3 things on 2 sites
  • Find what changed3 hours in charts, no datesCompare on: search ▼ 22%
  • Fix lost pagesguess redirects from memorycopy 14 lines from Not found
  • Did the fix work?no way to tellFixed: 0 not-found since Mon
  • Can AI read it?nobody askedone .md copy answers 404
  • Alert mail44 mails in one minutequiet: nothing open, no mail

6 jobs, all after the damage, Thursday to Sundaythe same 6 jobs, small, early, each with its proof

Sample week. The jobs are the same; what moves is when you learn about each one and how much digging it takes.

Why the work starts after the damage

Picture a common week on a small business site. A page breaks on a Monday, and nobody sees it. On Thursday a client writes to ask why leads are down, and someone spends an afternoon in charts working out which week, which page and which source. By then the damage is ten days old.

The charts were fine. They answered a question nobody had asked yet. Every one of those afternoons began with the same few worries, and each worry had an answer sitting in the server's access log the whole time: the file where your web server writes one line for every request it answers, from people, crawlers and bots alike.

So when we built the 2.35 and 2.36 releases of Skald, we started from the questions, written the way a site owner says them, and gave each one a place in the dashboard that already holds the answer. The hero above shows the same six jobs in a reactive week and a proactive one. The jobs do not go away. They get smaller and earlier, and each arrives with its proof.

Here are the twelve questions, grouped by the worry behind them.

Is it working? Four questions about this week

1. What is working? Overview › Summary has a card called "What changed: is it working?". It lists your sources and your landing pages, each with Now, Before and Change, and one sentence that names the biggest rise and the biggest drop. A landing page is the page a visit started on.

2. Which page is doing well? Traffic › Pages sorts by Rising and Falling: the pages with the biggest gain or drop against the period before, first. Both need Compare switched on, and say so when it is off.

3. Yesterday, 7 days or 30 days? And does that include today? Under the title sit two switches, Include today and Compare with the period before, and one line with the exact dates. On 15 January, "Last 7 days" with today is 9 to 15 January against 2 to 8 January; with whole days only, it is 8 to 14 January against 1 to 7 January. While today is included, Compare stays off and says why: today is not complete, so it cannot be compared fairly with a whole week.

4. How much traffic, and from which source? The key figures are five tiles: People, From search engines, Returning people, Leads, and Sent by AI assistants. With Compare on, each shows its change, as ▼ 40% · before: 125, or ≈ about the same when the gap is 3 or fewer. When the earlier period was never measured, for example before the site was added, the tile says "before: not measured" instead of showing a rise from nothing.

A tool that draws a missing week as zero will show you a rise from nothing the first week you use it. In Skald a figure that was not recorded is blank, never 0. The full tour of these cards is on the what changed page.

Work on the page, or on getting more traffic?

5. Should I work on conversion or on traffic? Conversion work (CRO, making more of the visits you have convert) and traffic work compete for the same hours, and the choice is often made by whoever argues loudest. The Summary card "Work on the page, or on traffic?" answers it from two lists.

  • Improve the page: pages with at least 10 visits that started there, where 90% or more left at once and no form was sent. The traffic is there; the page loses it.
  • Work on both: pages that search engines send at least 3 visits to, where 80% or more still leave at once. Search finds the page, and the page does not hold the people it brings.

"Left at once" has one meaning here: the visit loaded that one page and nothing else, as the server's log saw it. The card and the help guide both say so, because a number with a vague name gets argued about. Traffic › Pages has the same idea as a sort chip, Needs work.

Which new pages started getting traffic?

6. I published three pages last month. Is anyone finding them? The card "New and updated pages: are they getting found?" lists every page changed in the last 30 days, with its visits and search visits since the change, the first search visit, and whether Googlebot fetched it.

The change date comes from your website's own page list when Skald has one, and otherwise from the <lastmod> line in your sitemap, the date your site says a page last changed. Some site builders stamp every page with the build time. When most of the sitemap's dates fall on one day, the card says they look like build times, so you do not read a deploy as a hundred edits.

What should I fix this week?

7. What should I fix? Overview › Insights now opens with To do now: what to fix this week, worst leak first. Each to-do carries its proof (a figure and its days), what to do, and how Skald will check it. The kinds are old page addresses still asked for, server errors, AI asking for Markdown copies of live pages that are missing, a sitemap address crawlers ask for that is not there, and a page nearly every visit leaves at once with no form sent. Each appears only past its threshold: old addresses need at least 10 asks between them, for example.

When you mark one done, Skald checks it again on every whole day after the mark with the same measure that raised it, and writes the answer plainly: "Fixed: 0 not-found answers since 25 Oct", or "Not yet". A to-do the second count disagrees with is left out and named, never shown on a guess. More on the to-do list from your server log.

8. Which old addresses are still asked for? After a redesign, old links live on in bookmarks, other sites and search results. Site › Not found starts with "Old addresses still asked for": every page address that people, search engines and AI asked for in the last 30 whole days and that now fails, most asked first, with scanner probes left out. Beside each one is the closest live page Skald found, graded Same page, Same topic, Loose or No good match, and the line to paste for nginx or Apache: a 301 redirect, or 410 when the page is gone for good. Skald writes the line; you put it on your server. See 404 redirects from the log.

Did my change work?

9. I rewrote the pricing page. Did it help? On Traffic › Pages, open the page and press + Mark a change: the day, what changed (content, title, a link to it, a redirect, other) and a short note. Skald then sets the same number of whole days before and after side by side, up to 28 on each side, with the day of the change in neither.

It compares visits that started on the page, leads, search visits, AI visits, AI reads and the share who left at once. Each gets a verdict: Up, Down, No clear change (when the difference sits within two standard errors, the page's usual day-to-day wobble), or Too few to tell. With fewer than 14 whole days after the change it says Too early to tell, how many days it has, and the day the verdict will come. That waiting is the honest part. The details are on the did my change work page.

Can AI read my page?

10. When ChatGPT or Perplexity reads my page, what does it get? A page's story on Traffic › Pages now has an AI readiness block: six lines, each Good, Needs attention or Not known, with one line of evidence. Does robots.txt let each of 9 named AI crawlers read this path? Did crawlers meet errors or slow answers in the last 30 days? Is the page in the sitemap? Did the Markdown copy AI asked for answer? Did AI ask for an address that does not exist? When did AI last read it?

It is built from the log and Skald's own copies of your robots.txt and sitemap. Where Skald has no copy, the line says Not known instead of guessing. The full walk-through is on the AI readiness check, and the Markdown side of it in the note on Markdown copies.

Are my numbers true?

11. Can I trust the people count? Every question above rests on one count: which requests were people. 2.35 and 2.36 closed several gaps where a tool in a browser's clothes was counted as a person. A link checker that sends HEAD requests (a request for the headers only, without the page) for two or more pages, an address that asks for five or more probe paths the site does not have, an attack whose dots and slashes are encoded twice: each is now filed under bots, whole visit and all. A scanner that writes a crawler's name in its user agent is no longer believed as that crawler. And machine addresses (Markdown copies, WordPress's API, feeds) and an HTML sitemap page stopped counting as page views.

On skaldanalytics.com itself, over the days its log still held, the machine-address rule took page views across all groups from 6,426 to 5,551; 874 of the 875 views it removed were .md fetches, 164 of them by AI assistants, and people's own page views did not move. Separately, the estimate "likely sent by an AI assistant" fell from 54 visits to 1 for 18 to 24 October once chance was taken out, most of them on our free check page. Both figures are in Skald's 2.35.0 release notes (25 Oct 2026). Then each day is counted a second time by separate code, and a day the two counts disagree on is withheld. The longer story is in bots that look human.

Will I drown in alert mail?

12. If Skald watches all of this, will it mail me all day? It did once, and that is why 2.36 exists. On 25 October one person on our own server got 44 security mails in a minute: the first run after a fix sent one mail per open item, per site, across 6 sites. Now each person gets one summary mail a day at 01:00, covering every site they look after, with a section per site that has something open, at most 10 lines per part, and no mail at all on a quiet day. The note on alert fatigue shows how it decides.

The whole question map in one table

Here are the twelve questions with the place that answers each and the address that opens it. Every state in the dashboard has its own address, so you can bookmark a question or send it to a colleague.

filter by worry

#The questionWhere Skald answers itThe address
1What is working?Overview › Summary, What changed (sources and landing pages, now against before)#overview/summary?cmp=prev
2Which page is doing well?Traffic › Pages, sorted Rising or Falling#traffic/pages?cmp=prev&psort=rising
3Yesterday, 7 days, 30 days? With today?The period buttons, Include today, Compare, the exact dates#overview/summary?range=7d&inc=1
4How much traffic, from which source?Key figures with their change, people per day, where people came from#overview/summary
5Conversion or more traffic?Work on the page, or on traffic?#traffic/pages?psort=work
6Which new pages started getting traffic?New and updated pages, from the sitemap's lastmod#traffic/pages?psort=changed
7What should I fix?Overview › Insights, To do now, re-checked daily once marked done#overview/insights?todo=lost-pages
8Which old addresses are still asked for?Site › Not found, with the closest live page and the server line#site/notfound
9Did my change work?Mark a change; the same days before and after#traffic/pages?page=/pricing/
10Can AI read my page?AI readiness in a page's story; Pages AI can't read well#ai/assistants?arw=copy
11Are my numbers true?Tools and scanners filed as bots, machine addresses out of page views, a second count#site/quality
12Will I drown in alert mail?One summary mail a day at 01:00 per person#threats/alerts

Each address follows your dashboard's own address, for example spcaeo.github.io/skald/#overview/summary?cmp=prev in the live demo with sample data.

A short Monday routine, step by step

Put together, the twelve answers make a short weekly routine per site. When nothing is wrong it is a few minutes of reading; when something is, you see it before a client does.

  1. Read the 01:00 summary mail, if one came. No mail means nothing is open.
  2. Open Overview › Summary, Last 7 days, Compare on. Read the five tiles and the "What changed" sentence.
  3. Open Overview › Insights. Work the top to-do; copy its lines or send it to the person who can fix it.
  4. For each page you changed last week, press + Mark a change, so the verdict is waiting in two weeks.
  5. Glance at Done, and did it work? for the fixes from earlier weeks.

None of it needs a script tag on your pages or a call to an AI service. It needs the log your server already writes. If you want to try it on your own site, see the plans and the 30-day trial, or open the live demo and follow the table above.

questions

Proactive website analytics FAQ

Straight answers, including what Skald does not do.

What is proactive website analytics?

It is analytics that brings you what changed and what to fix, each with its figure, its days and a next step, instead of waiting for you to notice a fall in a chart. In Skald it rests on the web server's own access log, read every day, and every figure is counted a second time by separate code.

Which question should I start with?

Start with "is it working?": open Overview › Summary for the last 7 days with Compare on. It puts each key figure beside the 7 days before, with the exact dates, and names the biggest rise and drop among your sources and landing pages.

Do I need to check the dashboard every day?

No. Each person on a site's team with security alerts on gets one summary mail a day at 01:00 when something is open: security items, the open to-dos and set-up problems, for every site they look after. No mail comes on a day with nothing open.

Can Skald tell me why traffic fell?

It shows where it fell: which sources and which landing pages, against the same number of days before, and which old addresses now answer not found. The why is yours to judge; Skald gives the figures and the days, and says not recorded where it has none.

Does any of this ask an AI model?

No. Every answer on this page is read from your server's log, Skald's own copies of your robots.txt and sitemap, and the leads your tools send. Nothing is sent to an AI service to build it.

$ tail -f access.log | skald

See what AI does with your site. Free for 30 days.

No card. We set up hosted Skald for you within one working day, or you self-host it free. Every line of your log, sorted and added up.

Start free trial Live demo