Plugin Visibility Index · Run 03
Two AI engines give similar answers from almost none of the same sources
Asked the identical question in ten WordPress plugin categories, Perplexity and Google AI Mode named a median 53% of the same products — but cited a median of only 15% of the same domains. In one category they shared no source at all.
What was measured?
One prompt per category — “best WordPress [category] plugins in 2026” — put to both engines on 8 September 2026, within the same hour. For each answer I opened the full source list rather than reading inline citations, and recorded every domain in it.
That produced 106 source slots across 90 distinct domains for Perplexity, and 87 slots across 68 domains for Google. Between them, 130 distinct domains. Twenty-eight of those — 22% — were used by both engines anywhere in the corpus.
How much do the two source lists actually overlap?
| Category | Perplexity | Shared | Overlap | |
|---|---|---|---|---|
| Tables | 10 | 7 | 5 | 42% |
| Caching | 10 | 10 | 5 | 33% |
| LMS | 10 | 8 | 4 | 29% |
| Booking | 10 | 14 | 4 | 20% |
| Backup | 10 | 10 | 3 | 18% |
| Product feeds | 11 | 6 | 2 | 13% |
| Membership | 12 | 7 | 2 | 12% |
| Translation | 14 | 8 | 2 | 10% |
| Popups | 10 | 8 | 1 | 6% |
| Affiliate | 9 | 9 | 0 | 0% |
Set the two numbers side by side and the shape of the thing appears. On the same ten questions the engines agreed on 53% of the products and 15% of the sources. They are reading different corners of the web and arriving at roughly the same shortlist.
It means being cited is not the same thing as being recommended. Two largely disjoint sets of pages produced overlapping answers, which suggests the consensus about which plugins are good is thick enough that either engine finds it — and that a single page has less leverage over the outcome than its citation count implies.
Which channel is the biggest single difference?
YouTube. Google AI Mode cited youtube.com in eight of ten categories. Perplexity cited it in none.
| Source type | Perplexity | Google AI Mode |
|---|---|---|
| youtube.com, categories cited in | 0 of 10 | 8 of 10 |
| Vendor-owned domains, share of source slots | 41% | 47% |
| wordpress.org locale mirrors cited | 4 hosts | none |
This is the clearest example of something a vendor can act on. A review video is worth something on one engine and nothing on the other, and no amount of writing will substitute for it on Google. It also lines up with the wider evidence that YouTube presence correlates with AI citation more strongly than backlinks do — but here the effect is entirely on one side of the split.
Do both engines lean on vendor blogs?
Yes, and almost identically. 41% of Perplexity’s source slots and 47% of Google’s pointed at a domain owned by a WordPress product vendor. Independent review sites, hosting companies and personal blogs make up the rest.
So the engines disagree about which vendor blogs to read while agreeing that vendor blogs are roughly half the answer. For a plugin company that is the encouraging half of this run: your own writing is eligible on both engines. It is the specific pages that do not transfer.
Does one publisher show up on both engines?
One does, and it is the same network that dominated Attribution 01. FS Code publishes Booknetic (appointment booking) and FS Poster (social auto-posting); its three domains are fs-code.com, booknetic.com and fs-poster.com. I confirmed the ownership on their own site rather than inferring it from the names.
| FS Code network | Categories cited in | Of which it sells nothing in |
|---|---|---|
| Perplexity | 7 of 10 | 6 |
| Google AI Mode | 8 of 10 | 7 |
| Cited on both engines | 7 of 10 | 6 |
Backup, caching, LMS, popups, tables, translation — one booking-software company is a source on both engines in all of them. In a corpus where the two engines agree on only 15% of sources, this network is in the small set they agree on.
Whatever they are doing is the one thing in this data that demonstrably travels across engines. It is not a backlink strategy and it is not schema. It is a large volume of long, structured, comparison-shaped articles about categories next to their own.
Did the locale-mirror problem show up again?
On Perplexity, yes — in a fresh run, in categories where I had not seen it before. The product-feed answer cited ps.wordpress.org (Pashto) and yor.wordpress.org (Yoruba) alongside the canonical wordpress.org. The affiliate answer cited ast.wordpress.org (Asturian) and fur.wordpress.org (Friulian), again alongside the canonical.
Google AI Mode cited only wordpress.org. So the hreflang-splitting described in Attribution 01 looks like a Perplexity retrieval behaviour rather than something inherent to AI search. Four new mirror hosts in two fresh categories is a replication, not a one-off.
That last sentence was wrong. Run 05 put the same ten categories to ChatGPT and found locale mirrors in four of ten categories across six mirror hosts — twice as often as Perplexity. It is a behaviour of retrieval-based engines generally, absent only on Google, not something specific to Perplexity. The rest of this run stands.
What can this run not tell you?
One prompt per category, one run each, two engines, one day. Ten categories is a small base for a median, and the per-category rows are single observations.
The two engines expose their sources differently. Perplexity gives a numbered list that matches its stated source count. Google shows a panel of results alongside the answer, which is the best available view of what it drew on but is not guaranteed to be exactly the retrieval set. The comparison is between what each engine shows a user, which is the thing a vendor can check — not between two internal indexes I cannot see.
The vendor-owned share depends on my classification of 130 domains, and a handful are judgement calls. The 41%/47% figures should be read as “both engines take roughly half their sources from vendors”, not as precise to the point.
Run 01 measured Perplexity against itself on identical prompts and found 75% product overlap. There is no equivalent within-engine baseline for source lists, so I cannot say how much of the 15% gap is engine difference and how much is ordinary run-to-run churn in retrieval. That is the next run.
Which sources are speaking for your category?
Send me your wordpress.org slug. I will pull the full source list from both engines for your category and tell you who is being read — including the competitors and the outsiders. Free, no call, usually within a day.
Run 04 supplies the within-engine baseline this run was missing — and finds an identical query returns an identical source list.
Newer in this index
Run 04 supplies the baseline this page was missing. Repeating an identical query returns an identical source list — a cache. Change one character and a fifth of the sources move. Change engine and seven-eighths do. The cross-engine gap is about four times the noise inside either engine.