Just 10 Blog Posts and Still Cited by Perplexity: Anatomy of a 30-Day Citation Board

What This Piece Does

When AI answers a question, it often attaches sources. Which sites make that list? Opinions are everywhere; measurements are rare. Our method is blunt: we opened our own 30-day probe log, recorded the cited domain every time an AI attached a source, and ranked the results into a board. This is a dissection, not a census.

Limitations First

This board has three built-in boundaries. We state them before showing the numbers.

First, we wrote the questions ourselves. The probes use our own question set, so what the AI answers, and whom it cites, follows our questions. A different question set could produce a different board.

Second, the window is only 30 days. This is a short-term slice, not a long-term trend.

Third, the instrument is young. Our self-built observatory on km.idaeo.ai only came online at 2026-08-10 07:52 UTC. The sample is young, and we say so plainly.

So read the numbers below as "dissecting one specimen," not "measuring the whole world."

The Board Looks Like This

From the full history of the 30-day probe log, after removing 164 entries from Google's redirect shell vertexaisearch, the machine tally of cited domains reads:

• tradingeconomics.com: 27 citations
• youtube.com: 26
• prtimes.jp: 26
• cna.com.tw: 24
• idaeo family: 0

The last line is us. Zero is zero, and we print it as is. Own the ledger first, and the road stays straight. That is also why this series exists: measure where you stand before discussing where to go.

The Top Spots Come in Only Four Shapes

Line up the top names and only four shapes appear: data sites, video, press-release platforms, and wire services.

An everyday analogy: AI is a reporter on deadline holding a shortlist. For numbers, the reporter goes to a data site; for footage, to video; for corporate updates, to a press-release platform; for fact checks, to a wire service. Shops outside the list get walked past.

The four types share one trait: each supplies material that can be quoted directly and traced back to its origin. Data sites provide measured numbers, wire services provide reported facts, and press-release platforms provide statements in the subjects' own words. That is exactly the material an AI wants when assembling an answer — usable as is, with a source that points back.

So the first step toward being cited is not making the shop bigger. It is asking: when the reporter runs short of which kind of material will they think of you?

Small Shops Make the List Too

Do small sites stand a chance? A sitemap check gave a counterexample.

aicw.io has 1,739 URLs in total, of which 1,448 are directory routes; the real blog holds only 10 posts — and it is still cited by Perplexity. A wide spread of directory pages props up the storefront while the actual content pages are a thin stack, yet that did not keep it off the list. The control case, mersel.ai, has 178 URLs with 83 English blog posts — a different route: a small site with a high share of content pages.

The shared conclusion: being cited does not depend on store size, and volume is not the ticket. A shop with 10 articles can still make the reporter's shortlist.

Three Traits You Can Copy

From the cited sites, we extract three copyable traits:

• First-party data: measured and computed by you, not retold from others. For retold material, the AI goes looking for the original source — not for you.
• One question per page: one page answers one question, no grab bags. A reporter on deadline wants "the answer to this question," not your entire inventory.
• An English version: English dominates the AI world. In an earlier piece we measured the language structure of our own visibility at 74.7% / 22.0% / 3.3% (https://km.idaeo.ai/ai/visibility-lesson-1). Without an English version, you are absent from the main arena.

Being Crawled Is Not Being Cited

Some will ask: AI crawls me every day — does that not mean it thinks highly of me?

An earlier piece did the ledger: over five months, roughly 30 million crawls bought only 476 human clicks (a visible lower bound) (https://km.idaeo.ai/ai/ai-crawler-shop-ledger). Crawling is stocking; citation is shelving. This board adds another cut: however diligently the bots crawl us, the idaeo family still sits at 0 on the 30-day board.

Indexing rhythms also differ by company. The earlier measurement: OAI-SearchBot's first crawl takes a median of 332.8 hours (13.9 days), while ClaudeBot takes only 4.2 hours (https://km.idaeo.ai/ai/ai-crawler-marketing). Stocking can be fast or slow, but fast stocking does not guarantee shelving. Fast delivery only means the warehouse receives often; whether the goods reach the shelf depends on the goods themselves.

What the Official Channels Say

There is also a front door. Perplexity has a public publisher contact, [email protected], with no ranking promises attached. Google states officially that no AI text files are needed.

Read together, the two statements are clear: there is no paid fast lane and no secret file format. If someone tries to sell you a "guaranteed AI citation" service, hold it against these two statements first. The list is earned with content, not bought with formats or money.

Three Actions to Close

Three things. Do them as written.

First, pick one question where you hold first-party data, and write it as a standalone page that answers only that question.

Second, build the English version of that one page first. Do not translate the whole site.

Third, set a reconciliation date: run the same question set again and see whether the board now includes you.

We will follow the same three steps ourselves. At the next reconciliation, whether the idaeo line is still 0 — we will print it as is.