How is this different from “getting into best-of articles”?
how to get into “best tools” articles is about getting into rankings other people wrote. This guide is the other side: you (or your industry) build an independent scored ranking page, and how to organise it. It also differs from How do you write an “X vs Y” comparison page that AI cites?: a comparison page puts two products side by side, a ranking page lists a whole field under one question, with inclusion and sorting rules.
How many questions should one ranking page target?
One. Each ranking page on Pagent answers one question a buyer asks; the title is the question and the inclusion rule is in the data. Read on 2026-11-05, there are 27 pages; four of them:
| Ranking page | The question it answers | Inclusion rule |
|---|---|---|
| best-ai-assistant-on-telegram | Which AI assistant works best inside Telegram? | Only assistants that work in Telegram and were tested on at least 5 of the 10 tests |
| ai-assistant-with-no-app-to-install | Which assistants work from the messaging app already on your phone, with no app to install? | Channel is iMessage, SMS, WhatsApp or Telegram, no app needed, tested on at least 5 tests |
| ai-assistant-to-book-flights | Which AI assistant can book flights for you? | Ranked on the results of the flight-booking test |
| ai-assistant-for-restaurant-reservations | Which AI assistant can really get you a restaurant table? | Ranked on the results of the restaurant-booking test |
The same assistants and the same tests, with a different inclusion rule, answer a different question. So each page has content the others do not, instead of being one page copied with a word swapped.
What does a ranking page need on it?
Taking Pagent Telegram as the example, opened and checked on 2026-11-05:
| Part | How this page does it | Why it is there |
|---|---|---|
| Date, count, version | “Updated Oct 27, 2026 · 10 assistants ranked”, plus Rubric v0.1 | Readers see how fresh the data is, how many were compared and which version of the rubric applies |
| One sentence for the inclusion and sorting rule | Only assistants that work in Telegram and were tested on at least 5 tests, ranked by their average over the tests taken | The basis for the ranking is on the page, so it can be checked |
| One column per test, plus an average | 13 columns: rank, assistant, 10 tests, average | People and AI can check item by item and lift out a single cell |
| Each label states its rule | “Top pick” is the highest average; “Fastest that's still good” is the quickest median finish time among those averaging 7 or more | A label has a definition and is not handed out casually |
| How to choose | Three tips: is the channel really supported, check permissions before linking accounts, compare reply times if you use it all day | Turns scores into a buyer's decision |
| Real user voices | Public posts from users | A second source of evidence beside the scores |
| FAQ and a way to ask for a test | Four questions; “Request a test” lets a vendor ask to be added | Buyers' questions written as Q&A, and vendors get a proper way in |
The method has its own page, Pagent's methodology page, checked on 2026-11-05: no score without a logged run; untested items show “—” and are never guessed; before averaging, each assistant gets 3 neutral scores of 5, so assistants with fewer tests are pulled towards 5; scores older than 90 days turn grey. Every ranking page links to it.
How should prefilled scores be handled?
This is the most important point here. Pagent's data covers 47 assistants (read 2026-11-05): 22 have scores from our own tests and 25 have prefilled starting scores. All score dates are 2026-10-27.
On the Telegram page above, 6 of the 10 rows carry prefilled scores, and the table does not mark which is which. So this guide quotes no rank or score from that page and only describes its structure. This is something Pagent's page still has to add, not something already done:
- Mark every row as “measured” or “prefilled” and say how many tests it took.
- Draw the title, description and labels such as “Top pick” only from measured rows.
- Explain in the method what prefilled scores are and why they exist.
On a ranking that does not mark this, neither readers nor AI can tell whether a score was measured or filled in, and citing it means vouching for it.
How do you build your own ranking page?
1. Pick a question buyers really ask
Use the way buyers ask it, for example “which X works best in Y”. One question per page.
2. Fix the inclusion and sorting rules
Write it as one sentence on the page: who gets in, how many tests at least, what it is sorted by. If the rule changes the ranks change, and the page should show that.
3. Draw conclusions only from scores you measured
Leave untested items blank, no numbers. If you must show prefilled starting scores, label them row by row and keep them out of labels such as “number one”.
4. Put a date and number of runs on every score
A style such as “tested 2026-10-27, Dots 28 runs”; see How do you write an “X vs Y” comparison page that AI cites?.
5. Give the method its own page and link every ranking to it
How scores are made, what is tested, when they go stale: write it once and cite it everywhere.
6. Give the vendors you rank a proper way in
For example “request a test”. This is more reliable than asking vendors to beg authors for a listing, and it keeps the page getting new content.
7. After a few weeks, check whether it is cited
Ask AI the buyer's question a few times and see whether your page appears among its sources. We have not counted how often Pagent pages are cited, so we give no such figure. For how to judge the signals, see What signals does AI actually look at when recommending brands?.
FAQ
What is the difference between a ranking page and a comparison page?
A comparison page puts two products side by side; a ranking page lists a whole field under one question, with inclusion and sorting rules. How to write the former is in the comparison-page guide.
Do I have to run the tests myself?
To say “who is better” you need your own test records. Leave untested items blank and do not fill in numbers.
Can prefilled scores go on a ranking page?
They can, but label them row by row and do not let them decide the number one spot, the labels or the description. Readers and AI must be able to tell which scores were measured.
Why should each ranking page target one question only?
One question per page, each with its own inclusion rule, keeps the content from repeating. Cramming several questions into one page leaves every part incomplete and hard to lift out on its own.
Does a ranking page guarantee AI citations?
No. We have not counted how often Pagent pages are cited. This guide only covers how to organise the page and promises no effect.
Sources
- Pagent · Best AI assistant on Telegram （The operator’s own site, 2026-11-05, updated 2026-10-27）
- Pagent · Methodology （The operator’s own site, 2026-11-05, Rubric v0.1）
- Pagent · Pagent score data (27 reports, 47 assistants, read 2026-11-05) （The operator’s own site, 2026-11-05）
- kagentx evaluations: the data on the case study pages; each example gives its date.