What 21,700+ AI citations actually taught me
Two sites I own. 3 months. 21,700+citations from AI answer engines. Almost everyone writing about AI visibility right now is working from theory. This is the data — including the part that makes me look bad.
Across two sites over three months, Microsoft Copilot generated 21,700+AI citations. The lesson wasn’t the size of the number — it was that citations concentrate on tools and purpose-built pages, follow a winner-take-most pattern per query, and do not, by themselves, produce traffic or customers. A citation count is a vanity metric unless it lands on pages a buyer actually reaches.
The setup
I run two sites. Same owner, same three months, same measurement. They produced almost identical citation totals and completely opposite results. That accident is the whole study.
| HomeCalc.ca | HamiTahm.com | |
|---|---|---|
| What it is | Canadian real-estate calculators. Purpose-built, brand new. | A personal blog that later became a consultancy site. |
| Age | under three months old | about 18 months old |
| AI citations | 14,600+ | 7,100+ |
| Pages earning them | 25+ | 6+ |
| Shape of the curve | near-zero → steep, sustained growth | flat across the entire window — no growth, ~91% concentrated on a single page |
Roughly the same number of citations. One is spread across 25+ pages and climbing. The other is dominated by one page and flat. Total citation count, on its own, tells you almost nothing.
Finding 1 — A new site is invisible for about three weeks, then it isn’t
HomeCalc was a brand-new domain, under three months old, in a market owned by banks and national brokerages. For roughly the first three weeks of the window it earned effectively nothing. Then it moved — fast — to a peak of 420+ citations a day.
If you launch and see nothing for a month, that is not necessarily failure. It may just be the lag. Most people quit inside that window.
Finding 2 — Tools get cited. Opinions don’t.
The most-cited pages on HomeCalc, in order:
Four of the top six are calculators; the other two are practical how-to guides. Not essays, not opinion pieces, not thought leadership. Every one of them either computes a number or walks through concrete steps a person actually needed.
An answer engine is trying to answer a question. A tool that produces the answer is more useful to it than a page that has feelings about the question.
Finding 3 — AI picks a source and stays loyal
“Citation share” is the percentage of all citations for a query that went to this one site:
| Query | Citations | Share of all citations |
|---|---|---|
| property transfer tax bc | 263 | 29% |
| land transfer tax ontario | 160 | 39% |
| best rent increase calculator | 123 | 21% |
| calculate land transfer tax ontario | 102 | 72% |
| mortgage affordability calculator (100K salary) | 86 | 63% |
On calculate land transfer tax ontario, HomeCalc is 72% of every citation the query produces. On the affordability query, 63%.
This does not behave like the ten blue links, where positions two through ten still get something. It behaves closer to winner-take-most. Being a source is worth very little. Being the source is worth nearly everything.
Finding 4 — A purpose-built site earns citations. A blog buys a lottery ticket.
Now the same measurement on my personal site, and this is where it stops being flattering:
Roughly 91% of this entire domain’s AI citations land on one old essay about the 10,000-hour rule. It was written years ago, it is about skill mastery, and it has nothing to do with what I sell.
The page I actually sell — my AI visibility services page — earned 12 citations in three months.
HomeCalc was built on purpose, for a topic, with tools that answer questions. It got broad, repeatable, growing citations. My blog got one accidental jackpot and a flat line. Same person. Same three months.
Finding 5 — The one nobody in this industry wants to publish
AI citations are not traffic. AI citations are not customers.
My most-cited page in the world — 6,500 AI citations — produced 24 clicks from Google over the same three months. Twenty-four clicks, from an essay about skill mastery that has nothing to do with what I sell.
The whole site earned 44 clicks in three months, at an average position of 42.8.
I could have led this article with “21,700+ AI citations!” and you would have been impressed. It would also have been close to meaningless.
A citation count is a vanity metric unless the citations land on pages that sell something, for queries a buyer actually types. Mine mostly didn’t. That is not a reason to ignore AI visibility — HomeCalc proves the opposite. It is a reason to be extremely suspicious of anyone, including me, who shows you a big citation number and stops there.
Ask the follow-up question: cited on which pages, for which queries, and did anyone buy anything?
What earned nothing
Also true, and rarely published:
- Opinion and mindset posts: essentially zero citations.
- Off-topic content: a post about diets earned 61 citations on a site that sells AI visibility consulting. Useless citations are worse than none — they teach the model the wrong thing about who you are.
- My own commercial pages: 12 citations. The thing I most wanted cited was cited least.
What I would tell you to do with this
- Build the tool, not the take. If your page computes, compares, or resolves something concrete, it is a candidate for citation. If it merely has a view, it probably isn’t.
- Chase the query, not the count. One citation on a query your buyer types beats a thousand on a query they never will.
- Give it three weeks before you judge it.
- Measure what happens after the citation. If it produces no clicks and no customers, you have a trophy, not a channel.
Method, and what this doesn’t prove
- Citation data: Bing Webmaster Tools → AI Performance (Microsoft Copilot and partners), April 25, 2026 – July 25, 2026. Click and position data: Google Search Console, same window.
- This measures Copilot and its partners — not ChatGPT, Perplexity, Gemini or Grok, none of which publish citation data to publishers. Behaviour may differ on those engines. Anyone claiming otherwise with confidence is guessing.
- n = 2 sites, one owner, one market. This is a strong signal, not a law. Both sites are mine, which means I had total control and no stakeholders — advantages your business will not have.
- I am publishing the outcomes, not the build method. The method is what I am paid for, and I would rather tell you that plainly than pretend otherwise.
Use this data. Please. If you are writing about AI search and need real numbers instead of speculation, take them — charts, tables, figures. All I ask is a link back to this page so people can check the source.
↓ Download the HomeCalc sample (CSV) — the top cited pages and queries for HomeCalc only, CC BY 4.0. For the full two-site comparison (both domains, methodology, limitations), see the published dataset. Want the full per-page export beyond that? Email me.
Questions about the method, or want the underlying screenshots? hami@hamitahm.com. I’ll send them.
The related work
The full HomeCalc write-up is here: how a brand-new Canadian site earned 14,600+ AI citations in 3 months.
Want to know what AI actually says about your business?
Not a score out of 100. A real check, by a person, on the queries your buyers actually type.
Run the free AI visibility check →