Why Does an Agency's GEO Report Show Citations That I Cannot Reproduce in Perplexity?
You usually cannot reproduce an agency's Perplexity citations exactly because Perplexity generates a fresh answer every time. The sources it cites vary between runs. They also vary with the search mode and model, your account and its settings, your location and language, the exact wording of the question and the date. Many tracking tools query Perplexity through its API, which can behave differently from the app you use. A single missing citation is therefore normal. What should hold up is the pattern: across several runs of the same question, the cited pages should appear at roughly the rate the report claims. Ask the agency for the exact prompts, dates, settings and shareable links or saved answers, then rerun a sample together. If the pattern never appears, question the method.
Why do Perplexity results differ from the agency's report?
| Cause | What differs | How to control it |
|---|---|---|
| Run-to-run variation | Different sources cited for the same question | Compare rates across several runs, not one answer |
| Mode and model | Standard search, deeper research modes and model choice retrieve differently | Match the mode and model the agency used |
| API vs app | Tools often use Perplexity's API, which may differ from the consumer app | Ask which the agency used |
| Account and settings | Logged in or out, personalisation, saved preferences, Spaces | Test logged out or in a clean account |
| Location and language | Local and language-specific results | Match the location and language |
| Wording | Small wording changes retrieve different pages | Use the exact prompt text |
| Date | New pages and index updates change sources | Compare runs close in time |
| Citation mapping errors | Occasional mismatches between claims and linked sources, including reported bugs in API responses | Check the saved answer, not just a count |
What should you ask the agency for?
- The exact prompt text for each reported citation.
- Date and time of each run.
- Mode, model and method: app or API, and which search mode.
- Location and language settings.
- Number of runs per prompt, and in how many the citation appeared.
- Evidence: a saved copy of the full answer with sources, or a share link to the Perplexity thread where the app was used.
- How citations were counted: linked source, mentioned brand, or both.
A good agency can supply these without difficulty. Reporting "cited in 7 of 10 runs" with saved answers is verifiable; a single screenshot is not.
How do you rerun the check fairly?
- Choose five to ten reported citations, including the most important ones.
- Match the conditions: same prompt, mode, model, location and language, logged out or in a clean account.
- Run each prompt five times, close to the agency's dates if possible.
- Count how often the reported page is cited.
- Compare rates, not single results. If the agency reports "cited in most runs" and you see it in 3 or 4 of 5, that is consistent. If you see it in none of 5 across several prompts, ask for an explanation.
Doing this in a shared session with the agency removes most disagreements quickly.
When is the report the problem?
Be sceptical when:
- only single screenshots are offered, with no run counts;
- prompts include your brand name, which makes citations much easier to obtain;
- the agency cannot say which mode, model or method was used;
- reported citations do not appear in any of your matched reruns, across several prompts;
- the "citation" is a link to a page that does not support the claim it is attached to.
That last point deserves attention on its own. A citation that does not support the answer is weak evidence, as discussed in whether to count a Perplexity citation if the linked page does not support the claim.
What does a trustworthy citation report look like?
- A fixed list of buyer prompts, shared with you.
- Engines, modes and methods stated for each.
- Several runs per prompt, with counts: "your pricing page was cited in 8 of 15 runs".
- Saved answers or share links for inspection.
- Separate figures for brand-name prompts and category prompts.
- Changes over time on the same prompts.
What such tools measure, and their limits, is covered in what GEO citation tracking tools measure.
What do people ask about reproducing Perplexity citations?
Is it normal not to see the same citation every time?
Yes. Even strong pages are cited in some runs and not others. Consistency across many runs is the realistic test.
Do Perplexity Pro and free accounts give different results?
They can, because modes and models available differ. Match the account type the agency used.
Can agencies fake citation reports?
Screenshots can be selected or edited. Saved answers with dates, share links and run counts make that much harder.
Should we stop using an agency whose citations we cannot reproduce?
Not immediately. First ask for the method and rerun together; lack of transparency after that is the real warning sign.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources