Skip to content
Growzo · Perplexity

Perplexity is the one engine that can prove it cited you.

Growzo probes Perplexity with sonar-small-online — a live web-search model — and parses the citations array its response actually returns.Those URLs are the ones the engine consulted for that answer, which makes them evidence rather than interpretation, and it is why Perplexity is the only engine counted in Growzo’s “LLMs citing you” total. ChatGPT, Claude and Gemini answer from training knowledge and return no citations array at all, so their source lists are synthesized from hostnames in the prose and labelled an inference. Growzo keeps the two apart in the data model, not in a footnote.
A tracked prompt is sent to Perplexity's sonar-small-online model, which runs a live web search; the response carries both the answer prose and a citations array; Growzo parses that array into the ordered list of URLs the engine consulted, records the brand's position within it, and writes one ledger row holding the full answer, the mention flag, the position, the sources, a sentiment score and the word associations.
The array is read, not interpreted. That single step is what separates a citation from a guess.
01 · The mechanism

What a parsed citation actually is.

Perplexity retrieves from the live web before it writes anything, and the response it returns carries two parts: the prose a reader sees, and a list of the sources that prose was built from. Growzo reads the second part directly. There is no regular expression over the answer, no guessing which domain in a sentence was a source rather than an aside, and no reconciliation step where a hostname mentioned in passing gets promoted to a citation. The array names the URLs; Growzo records them.

Perplexity probe — one tracked promptcited · position 2
Modelsonar-small-online
Groundinglive web search
citations[] returned5 URLs
Your URL in the arrayposition 2
Sources parsed, not inferredall 5
Answer text storedin full
Verified citationIllustrative probe. The parsing step is not illustrative.

Why the distinction is not pedantry

On a dashboard a synthesized source list and a parsed one render identically: a logo, a domain, a tidy row. One of them is a record of what an engine consulted and the other is an artefact of a hostname appearing in a sentence. Treating them as the same number inflates the count of engines citing you, and it inflates it in the direction every vendor is tempted to inflate it.

The whole answer is kept

Not a snippet, not a hash — the full text as it was returned. That is what lets a recommendation be argued with rather than obeyed: you can open the row that produced it, read the answer, and see which four hosts sat around you. A recommendation without its evidence is an instruction from a black box.

02 · Order is half the score

Cited last is not the same as cited.

The citations array is ordered, so your place in it is a fact rather than an estimate — and that place is half of Growzo’s AI Brand Score, which multiplies mean mention frequency by mean position-inverse over a rolling 14-day window. A tool that counts citations without weighting order will report that you are winning a question you are quietly losing, because the third source in a five-source answer is read far less than the first. Multiplying rather than adding is deliberate: a near-zero on either factor pulls the whole number down, which is the honest behaviour.

Source order — one answer5 sources
1 · a competitor's comparison pageposition 1
2 · youposition 2
3 · a review roundupposition 3
4 · a forum threadposition 4
5 · a directory listingposition 5
Illustrative order. Position-inverse is what the score consumes.

Share of voice, per host

Over a separate 30-day window Growzo aggregates which hosts appear as sources across your tracked prompts and what share each one holds. It is the map of who currently owns the answer you want, and on most commercial questions it is a shorter and stranger list than the search results for the same phrase — which is exactly why it is worth looking at on its own.

improve_position is its own recommendation

Growzo emits five kinds of GEO recommendation, and moving up an order you are already in is one of them — distinct from earning a first citation, because the work is different. It is usually the cheapest real gain available, since the engine has already decided you are relevant.

03 · The boundary

What a citation proves, and what it does not.

This is the part most AI-visibility marketing skips, and it is the part that decides whether a number is safe to put in a board deck. A parsed citation is strong evidence about one specific thing and no evidence at all about several adjacent things a reader might assume it covers. Both columns below are stated so nobody has to infer the second one from the absence of the first.

It proves

The engine consulted your URL

It is in the array the response returned — not a hostname found in a sentence.

Where you ranked among sources

The array is ordered, so position is a fact rather than an estimate.

Who was cited instead of you

Every other URL in the array, on every probe, for every tracked prompt.

How you were described

Sentiment and word associations are scored from the full answer text.

It does not prove

That a human read that answer

Answers are generated per query and per session. No tool can see the audience.

That the next answer will cite you

Retrieval is proprietary and varies by query. One probe is one moment.

That you rank in any search engine

Growzo's rank tracking is Brave only, and that is a separate measurement.

That ChatGPT or Gemini cited you

They return no citations array. Their sources are inferred and labelled.

The honest summary: Growzo can tell you that Perplexity consulted your page on a prompt you chose to track, and where you sat among the sources it used. It cannot tell you that a person read that answer, and it will not pretend to. How the AI Brand Score is computed sets out the arithmetic that turns those probes into one number.

04 · The work, not the promise

What actually changes whether you get cited.

Nothing on this page is a lever anyone can pull on your behalf, and the largest single cause of a missing citation is not writing quality at all — it is access. An engine that cannot fetch your page cannot cite it, and that failure is invisible in rank tracking because your rankings do not move. So the order below is deliberate: access first, because it is verifiable, then the passage-level work, which is not.

The leverWhy it matters for a cited answerWhere Growzo measures it
PerplexityBot accessIt is the crawler behind the index Perplexity cites from. A Disallow removes you from the citation list entirely.The 13-bot registry and robots.txt reader — see the crawler tier
Self-contained passagesAn engine quotes one passage and has to be correct on its own. GEO scoring looks for roughly 40 to 120 word passages that stand alone.The per-page GEO score, written on every save
Real numbers and named sourcesEvidence density, original-claim language, how many real statistics the page carries and how many sources it names.The same GEO score, broken into its components
A stable anchor to citeA positional id like section-3 drifts to different text when the page grows a section, and the cited deep link silently moves.AEO scoring, which flags positional anchors
A page that exists at allexpand_coverage exists for tracked prompts your published corpus does not answer. There is nothing to cite on a question you never wrote about.The GEO recommendation engine
Growzo never claims a crawler obeys anything — robots.txt is a voluntary protocol with no enforcement, and what a tool can report is what your file instructs. And none of these levers is a guarantee: they raise the legitimate probability that an engine which retrieves before it answers finds something worth quoting. Anyone promising a citation is describing a mechanism that does not exist.
1 of 4
Engines returning citations
Perplexity. The other three are training-knowledge probes, labelled as such.
6
Fields per probe
Answer, mention, position, sources, sentiment, word associations.
14 days
Brand Score window
Mean frequency times mean position-inverse, times 100.
₹0.20
Metered per probe
Checked against a hard cap before the call, metered in paise after.
05 · One engine, not eight

Everything this page is careful not to claim.

Growzo is open source, self-hosted and bring-your-own-key, so a buyer can read the prober and check every sentence above in an afternoon. That is the reason the limits are published rather than left to be discovered: an overclaim on inspectable software has a very short shelf life, and the list below is more useful than the claim it replaces.

One probe is one moment

Perplexity answers a question fresh each time, and its retrieval is proprietary. A probe records what happened on that run, which is why the score averages a 14-day window rather than reporting the most recent answer as a status.

Four engines, not eight

Copilot, Grok, You.com and Phind appear in an internal enum with no prober behind them, so Growzo tracks nothing about them. This page will keep saying four until there is a fifth measurement to report.

No AI Overview detection

Google AI Overview presence is hard-coded false in the rank worker as a deliberate placeholder, so Growzo reports nothing about it rather than a zero that would read as absence. Rank tracking itself is Brave only.

No audience measurement

An AI answer is generated for one person in one session and is not published at a URL anyone can count. Growzo measures whether you were cited on prompts you chose to track, never how many people saw it.

Spending has a ceiling

Every probe is checked against a hard money cap before it happens and metered in paise after, with a soft alert at 80% and a refusal at 100%. The published unit economics are computed from the same constants the ledger bills against.

Pre-1.0, and said so

Growzo is pre-1.0 software. What is finished is described as finished and what is not is named — including on the overview page, which carries a whole section on what the product does not do.

The counterpart to this page is the one about the engines that cannot do any of it: what a ChatGPT, Claude or Gemini probe measures when there is no citations array to read. And the four-engine matrix puts all of them in one table. Anything unclear is worth an email to contact.

06 · Questions

Perplexity citation questions, answered plainly.

How does Growzo track Perplexity citations?

It probes Perplexity with sonar-small-online, a live web-search model, and parses the citations array the response returns. Those are the URLs the engine actually consulted for that answer, so Growzo records them as cited rather than inferred, along with your position in the order they were returned. Perplexity is the only engine Growzo probes where this is possible.

What makes a Perplexity citation different from a ChatGPT mention?

A structural difference in the response, not a difference of degree. Perplexity retrieves from the live web before it answers and hands back the list of sources it used. ChatGPT, Claude and Gemini answer from training knowledge and return no such list, so any 'sources' shown for them have to be synthesized by extracting hostnames from the answer prose. Growzo labels the second kind as an inference and excludes it from the count of LLMs citing you.

Can Growzo guarantee I will get cited by Perplexity?

No. No tool can, and one that implies otherwise is selling something it cannot deliver. Perplexity's retrieval and ranking are proprietary and its answers vary by query and by session. What Growzo can do is measure whether you were cited on your tracked prompts, record which hosts were cited instead, and emit recommendations describing work a person does — earn a citation from a host that keeps appearing, or move up the order you are already in.

Why does position inside the citation list matter so much?

Because it is half the AI Brand Score. The score multiplies mean mention frequency by mean position-inverse, so being cited last in a five-source answer is worth far less than being cited first — which matches how the answers are read. A tool that counts citations without weighting order will tell you that you are winning while you sit at the bottom of every list.

Does blocking PerplexityBot affect this?

Yes, directly. PerplexityBot is the crawler behind the index Perplexity cites from, and it is one of the thirteen agents in Growzo's crawler registry. A robots.txt that disallows it removes you from Perplexity's citation list, and the symptom is invisible in ordinary rank tracking because your rankings do not move — the loss shows up only as citations that never arrive. Growzo reports what your file instructs each agent to do, and never claims a crawler obeys it.

What does Growzo store for each Perplexity probe?

Six things, the same six it stores on every engine: the full answer text, whether your brand was mentioned, your position among the sources, the ordered source list itself, a sentiment score for how you were described, and the word associations the engine reached for. Because the whole answer is kept, every recommendation can be traced back to the row that produced it.

Track the one citation that is actually a citation.

Growzo runs on its own application and its own sign-in, on your own API keys. Track the prompts you want to be the answer to, and read the array rather than the adjective.

Start freeTalk to us