On 15 October 2026 the census site changed how it can be read. Every site Palo holds now has a record of its own, the census can be searched, listed, ranked and walked from site to site, and every page is written to be read by AI assistants as well as by people. Hlahlatha, the separate explorer, was retired. How the census counts did not change; the one change to the figures that day, the removal of malformed entries, has its own update.
Every site has a record
Each site in the census has a page at /explore/ followed by its name, for
example iol.co.za. It starts with the answer: the
site's status when Palo last checked it, and how many sites in the census
link to it and from it. Below that is what the site declares about itself
(its title, description, language and schema.org types, quoted as the site
wrote them), its place in the link graph with a one-line definition of each
measure, the .za sites it links to and from, 50 at a time with a link to
the next 50, and the domains outside .za it links to. Each linked site is a
link to its own record, so the census can be followed from site to site.
A link at the top opens the site itself.
Looking up a .za name the census does not hold says so. If the name exists in DNS, Palo adds it to the queue, ahead of the names it found itself, and its record appears once a crawler has visited it.
Questions across the census
- Search finds sites by words in their name or in the title they declare.
- Sites by zone lists the sites under co.za, gov.za, ac.za and the other zones, which can be narrowed by status, declared language and schema.org type.
- Rankings list the sites in the link graph by PageRank, by PageRank without footer-only links, by betweenness and by links in.
- Paths give the shortest chain of links from one site to another, and back, following links in the direction they point.
For AI assistants
Assistants such as ChatGPT, Claude and Perplexity read the web with plain
requests, do not run JavaScript, and some will only open addresses they
have already been given. So every record, list, ranking, path, measure and
report is also available as Markdown (the same address with .md), each
page links to the next thing a reader would want, and
llms.txt describes everything an assistant can read, with
links to all of it. Text that sites declare about themselves is shown as
quoted data, kept to one line and escaped, so it cannot add links or
headings to the page an assistant reads, and is marked as the site's own
words.
Every crawler is welcome to read the census, from the largest AI companies to anyone's own script.
Ask AI
The Ask AI page has prompts for common questions: what the census records about a site, who links to it, how two sites connect, which sites are about a topic, the sites in a zone, the most linked-to sites, how the .za web has changed, and how the census is made. Each prompt gives an assistant the census pages that answer the question and asks it to say which analysis its figures come from; it does not tell the assistant what to conclude. Every site record and measure page has one too.
A prompt can be copied into any assistant, or opened directly in Claude, ChatGPT, Perplexity, Grok, Kimi or Qwen, the assistants that accepted a prompt in a link when tested on 15 October 2026. Gemini, DeepSeek and others can be used by copying. Palo counts how often each prompt is opened in each assistant, or copied, and records nothing else.
Hlahlatha retired
Hlahlatha, whose name is a Sepedi word meaning to roam about, was a separate explorer for the census, built with JavaScript. The census pages now answer what it answered, in a form people and assistants can both read, so its address now leads to them: a site it was showing opens that site's record. Its data interface was retired with it, as were its live visits to sites, which did not check a site's robots.txt, and its cached WHOIS registration data, which was deleted.
Sites built with JavaScript
Some sites send an almost empty page and build their content, links
included, in the browser. Palo's main crawler does not run JavaScript, so
their links are likely missed until Palo's browser crawler visits them.
The main crawler now recognises such pages from their HTML and marks them,
and the browser crawler, which now runs on a schedule, visits those
sites first. Before this, about a quarter of the browser crawler's visits to
active sites found links. As of the analysis on 15 October 2026 at 20:45
SAST, about 450 active sites were marked, a number expected to rise as
sites are crawled again. The count is in the census
summary (client_rendered) and the timeline
CSV, and is a minimum: not every such site is
recognised. The browser crawler also now waits at least a second between
pages on a site.
The charts
The charts on the front page and the measure pages can be narrowed to the last 7, 30 or 90 days, or to the days since the counting rules changed on 11 October 2026. The changes of 11 to 13 October are now shown as one event, the refactor, and the event markers sit above the charts so they no longer cover the data.