Perplexity cites your brand, Gemini ignores it, and your audit is working fine

Perplexity cites your brand. Gemini, asked the same question a minute later, doesn't mention you.

M
Mike O’Sullivan · 9 min read
Perplexity cites your brand, Gemini ignores it, and your audit is working fine

Summarize with AI

AI search engine citation differences show which page each engine trusted, and what to fix first.

Perplexity cites your brand. Gemini, asked the same question a minute later, doesn’t mention you.

Somebody on the team says the audit must be broken. I’d say it has just handed you its most useful result.

AI search engine citation differences are normal, and they tell you more than a clean sweep would. Each engine runs its own searches, pulls its own sources, and backs its own claims. When two of them disagree about you, they’re showing you which of your pages carries real evidence and which one is thin.

The question worth asking is which page each engine trusted, and for which claim.

Why the same question gets two different answers

Each engine does its own research before it answers.

OpenAI’s ChatGPT search documentation says it typically rewrites your prompt into one or more search queries of its own. Google’s guidance on AI features in Search describes a “query fan-out” technique, where one question can become several related searches across subtopics. It also says AI Mode and AI Overviews can show different links from each other.

If two features inside Google can disagree, it’s no surprise when two separate companies do.

So a citation is a narrow thing. A citation shows that one system used one source in one response. It doesn’t prove your brand owns the topic across AI search, and it doesn’t predict the next answer a shopper sees.

Say one of your products is a merino wool sweater, and the question is where to find one for travel. Perplexity might cite your collection page, because it lists the product details for each style. Gemini might favor your packing guide, because it explains how folding and layering prevent wrinkles.

A third engine might skip you for a publisher’s roundup of travel knitwear. The publisher made the recommendation, and your site offered no direct proof for the claim the engine chose to make.

That’s three sources for one question, and each answer is a fair reading of the evidence that engine found. Compare them side by side.

PlatformWhat to compare for the same shopper promptWhat to record
PerplexityWhether it names and links to a product, guide, publisher, or retailer sourceAnswer wording, cited URLs, and the claim each URL supports
GeminiWhether it favors product evidence, editorial guidance, or another source typeBrand mention, supporting source, and any missing product detail
ChatGPTWhether its search response uses your page or a competing source for the recommendationRaw response, linked sources, and check date
Google AI ModeWhether its response surfaces a discovery page, care guide, or publisher resultVisible links, source category, and the shopper claim answered

Run every weekly audit as the same checklist:

  • Use the exact shopper prompt and record the platform and check date.
  • Save the raw response, including every cited or linked URL.
  • Open each cited source and read the claim beside the citation.
  • Compare that evidence with the page you wanted the engine to use.
  • Name the gap before you change any copy: a missing guide, a missing material detail, missing product proof, or a source format the engine preferred.

The last step saves the most work. Without it, an absence turns into a rushed rewrite of your collection copy when the real gap was a guide you never wrote.

How to read AI search engine citation differences

An ivory satin slip dress hangs on a clothing rack in a warmly lit room.

Being named and being cited are separate results.

Your brand can appear in a recommendation while someone else’s page supplies the facts that make the answer useful. A brand mention and a supporting citation are different findings, and they lead to different work.

Take a shopper who asks where to buy a satin slip dress, then asks how to wash satin. If Perplexity cites your collection page for the first question, that supports product discovery, because the page helps someone compare styles and shop. If Gemini cites your fabric-care guide for the second, that supports a care answer, because the guide explains washing in language the response can use.

Neither page failed. Each answer had a different job, so judge each page against its job. I’d drop the hunt for one universal visibility score.

Audit fieldWhat to recordWhat it helps you decide
Prompt intentBuy a satin slip dressWhether you need a discovery page
Cited pageYour collection or a rival URLWho supplied the evidence
Source typeBrand page, publisher, retailer, or forumWhich format won
Claim supportedStyle choice or wash adviceWhat the response needed
Missing evidenceCare detail or product proofWhat to add or improve

Record who was cited in your place before you blame your own content. A missing citation can come from a weak page. It can also come from a page the engine couldn’t retrieve, or from an answer built around a source type you don’t publish, such as a publisher roundup or a retailer listing.

I’d rather audit ten high-value shopper questions again and again than collect hundreds of broad prompts nobody can act on. Start with the questions that shape a purchase or remove a real objection, such as whether the dress works for a wedding guest outfit or how the fabric should be washed after wear.

Do that and the screenshots turn into a decision list. A collection may need clearer product detail. A care guide may need a direct link to the dresses it describes, so a factual answer can also help a shopper reach a purchase.

Separate evidence from volatility

A person measures the hem of a dusty-rose midi dress with a yellow measuring tape on a worktable.
A mauve formal dress on a mannequin is displayed beside a window, with its skirt lifted to reveal the lining.

Generative search changes between checks. The engine retrieves different sources, reads the prompt a little differently, or words the answer another way. That’s normal behavior, and it means a single audit run can’t carry a content decision by itself.

Repeated appearances are stronger evidence than a one-time citation. Keep a small, fixed set of shopper prompts, use the same wording each time, and look for the pattern across checks.

Say your midi dress product page appears once for an occasionwear question. In later checks, fashion magazine guides keep showing up for fit and length advice. That pattern tells you something specific: your brand has some recognition, and your content doesn’t yet carry the factual claim shoppers need.

It’s a more useful note than “Gemini ignored us.” The magazine guide may explain where the hem sits and how the cut works on different heights, while your listing offers photos and a size selector.

Before rewriting any copy, I’d open the page that was cited and note what it does. Does it answer the shopper’s question directly? Does it name the relevant product details? Can search crawlers reach it? Then I’d hold my own page to the same three tests.

A page can lose simply because it never states the answer in words. A listing that says “occasionwear” is thin evidence for a question about length or fit. A short fit section with the garment length and model details gives an AI response something concrete to cite.

Repeated checks take time and disciplined prompt records. One-off reviews cost more, because they create false confidence and send teams into content work the evidence never asked for.

So I’d treat a recurring publisher citation as a content brief. Build the missing explanation on your own site, connect it to the relevant products, and run the same prompts again once the page is live.

Build pages engines can verify

Citation differences often expose a simpler problem: your site lacks clear proof for the shopper question in front of it, so focus on pages that give search systems specific details they can retrieve, connect to the right product, and use with confidence.

Clear product evidence gives search systems material they can retrieve and support. Assign each high-value shopper question to the page that owns it, so when someone needs to know whether a dress suits a formal evening wedding, the relevant product page should state its fabric, explain the fit and length, and provide care instructions and available sizes when those details are true for that garment.

I’d fix the page closest to the shopper’s question before commissioning broad brand content, because “An elegant occasion dress” leaves too much work for everyone reading it, while an explanation of the lining and shoulder-to-hem measurement gives a shopper useful facts and details about waist fit give search systems solid material to reference.

Editorial pages earn their place when they answer the broader choice behind a purchase, so your wedding guest dress collection should link to a guide that explains dress codes, sleeve coverage for a religious ceremony, and dress lengths that work for a garden reception. That guide should send readers to relevant dresses rather than leave them with advice and no sensible next step.

Use audit notes as an evidence-gap list, and when answers repeatedly lean on outside sources because someone asks how to wash knitwear, build or improve a care guide with instructions that match the fibers and construction of the garments you sell, covering real decisions including whether an item can go in a machine or needs to dry flat.

This work also strengthens the wider signals around your store. Kyoto Pearl recovered 100% of its pre-migration organic traffic and increased AI citations 5x after its site recovery work, which I’d take as a reminder to repair missing evidence before chasing more output because a well-supported page gives every search system more to work with.

Build pages engines can verify

Citation differences often expose a simpler problem: that your site lacks clear proof for the shopper question in front of it. The fix is a page that gives search systems specific details they can retrieve, connect to the right product, and use with confidence.

Clear product evidence gives search systems material they can retrieve and support. Assign each high-value shopper question to the page that owns it. If someone needs to know whether a dress suits a formal evening wedding, the product page should state its fabric, explain the fit and length, and give care instructions and available sizes, when those details are true for that garment.

Google’s guidance points the same way. It asks for important content to be available as text, and for internal links that help people and crawlers find it.

I’d fix the page closest to the shopper’s question before commissioning broad brand content. “An elegant occasion dress” leaves the reader to do all the work. A line on the lining and the shoulder-to-hem measurement gives a shopper useful facts, and a note on waist fit gives a search system solid material to reference.

Editorial pages earn their place when they answer the broader choice behind a purchase. Your wedding guest collection should link to a guide that explains dress codes, sleeve coverage for a religious ceremony, and lengths that work for a garden reception. That guide should send readers back to relevant dresses, so nobody is left with advice and no sensible next step.

Use your audit notes as an evidence-gap list. If answers keep leaning on outside sources whenever someone asks how to wash knitwear, build or improve a care guide whose instructions match the fibers and construction of the garments you sell. Cover the real decisions, including whether an item can go in a machine or needs to dry flat.

This work strengthens the wider signals around your site too. Kyoto Pearl recovered 100% of its pre-migration organic traffic and non-brand visibility within 90 days, while AI citations increased 5x. I read that as a reason to repair missing evidence before chasing more output.

So the next time two engines disagree about you, keep both answers. The one that left you out shows whose page did the job yours didn’t, and that’s your next brief.

Frequently asked questions

Why does one AI answer cite my brand while another leaves it out?

Different systems can cite different sources for the same shopping question, so a Perplexity citation and a Gemini omission record a real difference in that response. For a query such as “Where can I find a merino wool sweater for women?”, compare the pages each answer cites and the claims they support. Your womenswear content needs clear evidence about the sweater’s material and fit.

How often should I repeat an AI citation audit?

Run an AI citation audit on a fixed schedule and after you make a meaningful change to your content or product pages. Use the same shopper prompts each time, then record cited sources and missing claims. I’d focus on patterns across repeated checks because one answer only captures what that system could support when you ran the prompt.

Should I create a new article whenever my brand is missing?

No. Create a new article when repeated audits show an evidence gap your existing content can’t answer. If Gemini skips your womenswear brand for a merino wool sweater query, check whether your category or product content explains the material, fit, and care details a shopper needs. I’d improve that existing evidence before adding another article that covers the same question.

Sprite builds brand authority through continuous, automated improvement. Quietly. Consistently. And at Scale.

No commitment
30-day free trial
Cancel anytime
Your Turn

See What You Could Save

Discover your potential savings in time, cost, and effort with Sprite's automated SEO content platform.