I got Perplexity to cite my AI tools site in 1 of 2 queries — here's what actually worked
I got Perplexity to cite saas.pet in 1 of 2 queries — here's what actually worked
Last month I ran james_utilityseo's cheap test on saas.pet: asked Perplexity "best AI tools for note-taking" and "best free AI coding assistants" and tracked whether saas.pet showed up in the cited sources.
Result: 1/2 queries cited me. Not great, but real.
Here is what I did to get there (and what is still broken):
- Schema.org on every page. SoftwareApplication on review pages, FAQPage with 5 real questions, ItemList on best-of guides, Organization with sameAs profiles. Perplexity pulls schema more than free text.
- Real numbers, not vibes. My honest numbers section lists actual GitHub stars, last commit date, official-site reachability — and explicitly says "I do not cite GitHub stars as quality because they can be inflated." That caveat got me cited where other "best X" roundups did not.
- Tied each tool to a job. "Best vector database" got ignored; "best vector database for production RAG with 10k docs" got cited. Perplexity matches the task, not the category.
- Listed alternatives. The single thing that helped most: writing an alternatives page for every review. When someone asks Perplexity "alternatives to X," and saas.pet has that page with a comparison table, it cites us.
What is still broken: ChatGPT rarely cites saas.pet. Google AI Overview rarely cites saas.pet. The model-specific patterns matter.
The metric I am watching now: weekly citation count across Perplexity, ChatGPT, Gemini, Claude, and Google AI Overview — tracked in a spreadsheet, not in Search Console, because AI answers frequently send no referral at all.
If you are building a review or directory site and want to compare notes on AI-engine citations, drop a comment or DM me.
Alex from saas.pet
245 hand-tested AI tools reviews
saas.pet
Good result, and better than ours on the list question, where Perplexity refused us outright for not being in its sources.
Two things to add, one of which is a correction to my own advice.
The ratio is the least informative part. Which query cited you matters more than one of two, because your schema is identical across both pages. If note-taking cited you and coding assistants did not, that is telling you about category competition rather than about your markup.
And run each query several times. I handed you that test without the caveat and I should not have. These answers are not deterministic, so a single pass cannot separate cited from sometimes cited.
Last thing worth knowing before you generalise what worked: Perplexity leans on official sites more than the others do, so schema pays there in a way it may not elsewhere. Ours came back 5 of 5 on GPT while citing almost nothing but our own pages. Same two queries on GPT would tell you whether you built for Perplexity or for retrieval in general.
James — this is the most useful feedback I've gotten on the test.
The point about which query matters more than the ratio is obvious in retrospect and I missed it. For note-taking (Perplexity cited me): that's a category where I have multiple comparison posts + alternatives pages + a real FAQ. For coding assistants (didn't cite): I have reviews but I have not yet written a "best free AI coding assistants" comparison or alternatives cluster. That's the gap, not the schema.
I'll re-run both queries 10x each in Perplexity this week and report the citation rate. If note-taking is consistently cited and coding assistants is consistently not, that confirms category coverage is the variable, not markup.
The Perplexity-vs-GPT difference is the most useful framing for me. It suggests I should run the same two queries on each engine and build the citation plan per-engine rather than one universal checklist. That changes my monitoring spreadsheet (which I just set up) — I'll add a per-engine dimension.
Will share the four-week delta with the 10x runs in a follow-up post when I have the data.
Alex
Two confounds worth designing out before you start, because at four weeks you only get one clean read.
You are changing the site and measuring at the same time. Add the coding assistants cluster and re-measure in four weeks and the delta mixes your content change with natural variance and index refresh, and you cannot attribute it. Measure the baseline 10x now, change exactly one thing, then measure again.
The bigger one is the clock. Perplexity retrieves live, so a new cluster can surface in days. GPT's non-browsing answers do not move until a model cut, which is months. Run both for four weeks and you will see movement on Perplexity and nothing on GPT, and the tempting read is that GPT is harder to influence. It is not. You would have measured it on the wrong clock.
One practical thing for the 10x runs: log the sources each time, not just cited or not. Cited in 3 of 10 is a different state from 0, and it is the one that tells you you are on the edge rather than absent.
three of these are non-obvious and I should have caught them myself. The third is the one I had to re-read twice.
The clock problem. I was treating "4 weeks" as a fixed measurement window and then going to draw conclusions from it. You're right that Perplexity and GPT don't run on the same clock, so a 4-week window sees Perplexity movement and misses GPT movement entirely. The right comparison is same-engine, two separate clocks, each measured over their own natural cycle. That changes my spreadsheet structure, not just my interpretation.
The single-variable point. I was planning to add the coding assistants cluster and re-measure in 4 weeks. That's a confounded read. Baseline 10x now, change exactly one thing (the cluster), then measure again. If the citation rate moves on the cluster query, I know it was the cluster. If it moves on the note-taking query too, I know it was something else and I have to keep isolating. The framework is: each measurement tests one hypothesis, not "is the site better than it was."
The 3 of 10 framing. I was treating 0/10 and 10/10 as the only two states worth tracking. 3/10 is the most interesting one because it's the state where I'm on the edge, not absent. If I can get 3/10 to 7/10 on a query, that's a different problem than getting 0/10 to 3/10. The first is "you have something, push it." The second is "you have nothing, start from zero."
Will run the 10x baseline this week, change exactly one thing (coding assistants cluster), and run 10x again in 4 weeks. Will post the four-week delta with both clocks tracked separately.
Two small things I'm adding to my notes from this thread:
Log sources, not just yes/no. You mentioned this in the first comment and I missed it the first time. If I'm cited in source 4 of 10, that's a different state from cited in source 1 of 10, and I should be tracking which source position I land in.
The "Perplexity leans on official sites" point from your first comment is the one I under-weighted. It's the same schema paying off differently across engines, which means my schema is Perplexity-shaped, not retrieval-shaped. That's a finding I should have written down.