Audit Perplexity 2026: Checklist untuk Brand dan Publisher

Audit Perplexity yang bagus bukan pencarian nama brand lalu screenshot.

Ia harus menjawab lima hal:

Apakah brand ditemukan?

Apakah sumber yang benar digunakan?

Apakah answer akurat?

Apakah visibility terjadi pada query yang bernilai?

Apakah ada dampak yang bisa diukur?

Perplexity pada 2026 adalah search environment yang mencari web secara real-time, menampilkan citations ke sources, punya Pro/research experiences, Search API dengan ranked results dan filters, serta premium data integrations pada use case tertentu.

Itu membuat audit lebih kompleks daripada rank tracking.

Tidak ada satu posisi.

Satu query bisa punya banyak sources.

Source bisa berubah.

Mode bisa berubah.

Brand bisa disebut tanpa owned citation.

Citation bisa ada tetapi claim salah.

Maka audit perlu checklist yang memisahkan technical, source, semantic, reputation, dan business layers.

Bagian A: Scope

1. Tentukan objective.

Brand accuracy?

Category discovery?

B2B shortlist?

Publisher citation?

Product discovery?

Reputation?

Jangan audit semua tanpa prioritas.

2. Tentukan surface.

Perplexity default.

Pro Search.

Research mode.

Search API.

Jangan campur consumer UI dan API.

3. Tentukan geography.

Indonesia.

Jakarta.

Global.

Perplexity Search supports location/region filters in API context.

4. Tentukan language.

Bahasa Indonesia.

English.

Perplexity Search supports Indonesian language filtering.

5. Tentukan period.

Tanggal audit.

Repeat cadence.

Current topics need more frequent.

Bagian B: Query Panel

6. Build query panel dari buyer/user questions.

Sales.

Search.

FAQ.

Support.

Research.

7. Pisahkan branded dan non-branded.

Branded = accuracy.

Non-branded = discovery.

8. Cluster.

Awareness.

Category.

Comparison.

Recommendation.

Procurement.

Risk.

Implementation.

Local.

9. Lock core panel.

Version.

No cherry picking.

10. Add exploration panel separately.

New trends.

No impact core denominator.

Bagian C: Sampling

11. Use multiple runs.

Three per query practical.

12. Record exact query.

No paraphrase mid-test.

13. Record model/mode if visible.

Perplexity model options can change.

14. Record timestamp.

WIB/UTC clear.

15. Define valid observation before testing.

System error excluded by rule.

Brand absent not excluded.

Bagian D: Brand Presence

16. Track brand mention.

Yes/no.

17. Track context.

Listed.

Recommended.

Compared.

Critical.

Historical.

18. Track recommendation separately.

Mention isn't recommendation.

19. Track product/service entity separately if needed.

Parent brand vs product.

20. Track entity confusion.

Namesakes.

subsidiaries.

old names.

Bagian E: Citation

21. Track owned-domain citation.

Exact URL.

22. Track third-party source about brand.

23. Track competitor sources.

24. Track source type.

Official.

publisher.

community.

government.

academic.

premium data.

directory.

marketplace.

25. Track source role.

Definition.

Fact.

spec.

comparison.

recommendation.

risk.

original research.

26. Track source currentness.

Publication.

last update.

fact date.

27. Track source accessibility.

Public.

paywall.

premium integrated.

auth.

28. Track source label if Perplexity shows one.

But remember Perplexity says source labels do not guarantee article accuracy.

Bagian F: Accuracy

29. Break answer into important factual claims.

30. Map citation to claim.

31. Verify support.

Supported.

Partial.

Unsupported.

32. Verify entity.

Correct company/product.

33. Verify units.

Currency.

percentage.

million/billion.

34. Verify date.

Current vs historical.

35. Verify scope.

Indonesia vs global.

SMB vs enterprise.

36. Flag high-risk errors.

Price.

CEO.

location.

security certification.

regulation.

availability.

37. Use primary sources for high-stakes validation.

Perplexity is research assistant.

Not final legal/medical/financial authority.

Bagian G: Hallucination Risk

38. Check inaccessible source situations.

Perplexity developer guidance warns paywalled/private/auth sources can increase hallucination risk.

39. Check very recent information.

Web may not have stabilized.

40. Check repeated related-but-not-matching search results.

Another documented hallucination risk.

41. Don't trust typed URLs generated by model text.

Use structured citation/source links.

42. For ambiguous query, add specific context in separate diagnostic test.

Entity.

date.

location.

Bagian H: Freshness

43. Classify topic volatility.

Real-time.

fast.

medium.

evergreen.

historical.

44. Review old citations on volatile facts.

45. Don't punish old source on evergreen topic.

46. Compare published vs updated date.

47. Avoid fake date refresh.

48. If using API research, test recency/date filters separately.

Label as controlled experiment, not consumer default.

Bagian I: Source Preference

49. Test meaningful query variations.

Intent.

language.

location.

buyer segment.

risk.

50. Change one variable at a time when possible.

51. Record source-set changes.

52. Don't call one source “preferred” universally.

Preference is query-conditioned.

53. If using domain allowlist/denylist, record.

Filtered test isn't organic visibility.

54. Compare source categories, not only domains.

Sometimes regulator should win.

Bagian J: Citation Drift

55. Track sources over multiple periods.

56. Record new/lost source.

57. Track persistence.

58. Distinguish expected high-drift query from evergreen.

59. Don't rewrite after one lost citation.

60. Use no-change window to understand natural variability.

61. Keep platform event log.

Model updates.

premium sources.

feature changes.

Bagian K: Website Source Quality

62. Critical facts public?

63. Pricing clear?

64. Product/service capability specific?

65. Leadership current?

66. Location/service area current?

67. Security/procurement info available where appropriate?

68. Original evidence?

69. Methodology?

70. Stable URLs?

71. No broken pages?

72. Readable text, not image-only?

73. Clear dates/version?

74. No duplicate content factory?

75. Content useful without Perplexity context?

If page only exists to “rank in Perplexity,” question it.

Bagian L: Schema and Structured Data

76. Use valid structured data for legitimate machine-readable/web purposes.

77. Match visible content.

78. Don't claim Schema.org is a documented direct Perplexity citation factor.

79. Don't confuse Perplexity API JSON Schema output with webpage Schema.org.

80. Measure citation separately.

No superstition.

Bagian M: Reputation

81. Audit third-party sources.

Media.

reviews.

community.

directories.

premium databases.

82. Negative sources accurate?

83. Old complaints resolved?

84. Company profiles current?

85. Earned coverage exists?

86. No fake reviews or synthetic authority networks.

87. Separate opinion from factual error.

88. Route corrections to source owner.

Website fix isn't universal.

Bagian N: Indonesia Context

89. Test Bahasa Indonesia.

90. Test English if audience international.

91. Look for local primary sources.

Government.

regulators.

BPS.

industry.

92. Flag when answer relies only on global sources for Indonesia-specific claim.

93. Measure local source scarcity.

94. Publish original local data if you genuinely have it.

95. State sample limitations.

Don't turn Jakarta sample into Indonesia claim.

Bagian O: Marketing Value

96. Define high-value query clusters.

97. Citation frequency on fixed denominator.

98. Brand mention frequency.

99. Recommendation frequency.

100. Accuracy rate.

101. High-relevance citation frequency.

102. Third-party source mix.

103. Referral from Perplexity if analytics can identify it.

104. Branded search as supporting signal.

105. Self-reported AI discovery.

106. CRM/pipeline.

107. Publisher subscription/engagement.

Do not collapse all into one score.

Bagian P: KPI Decision

108. Is Perplexity material to buyer journey?

109. Is methodology stable?

110. Is query panel business relevant?

111. Is owner assigned?

112. Does metric trigger action?

113. Is accuracy guardrail defined?

If yes:
citation can be KPI.

If no:
keep observational.

Bagian Q: Competitive Analysis

114. Pre-register competitor set.

115. Track competitor mentions.

116. Track competitor owned sources.

117. Open cited pages.

118. Identify information advantage.

119. Don't copy structure blindly.

120. If competitor has original data, create your own evidence, not duplicate.

121. If competitor wins because fit better, accept.

GEO isn't forcing bad recommendation.

Bagian R: Publisher Checklist

122. Which articles are repeatedly cited?

123. Original reporting or summary?

124. Citation role?

125. Referral?

126. Engagement?

127. Subscription?

128. Old pages accurate?

129. Paywall strategy intentional?

130. Public abstract available?

131. Corrections visible?

132. Stable archive?

Publisher value = influence + audience + economics.

Bagian S: B2B Brand Checklist

133. Category query presence.

134. Shortlist.

135. Comparison.

136. Procurement.

137. Security.

138. Implementation.

139. Pricing model.

140. Industry fit.

141. Risk.

142. Third-party evidence.

143. Premium company database accuracy.

B2B visibility should map deal stages.

Bagian T: Ecommerce Checklist

144. Product facts current.

145. Price.

146. availability.

147. variant.

148. local market.

149. warranty.

150. review sources.

151. retailer sources.

152. product mention vs citation.

153. purchase/referral outcomes.

No citation KPI if product isn't actually available.

Bagian U: Data Structure

154. One row per observation.

155. Unique observation ID.

156. Query ID.

157. panel version.

158. run number.

159. source URLs.

160. source role.

161. accuracy.

162. business value.

163. notes.

164. raw evidence reference.

At scale:
separate source table.

Bagian V: Quality Control

165. Second reviewer sample.

166. Classification rubric.

167. No deletion of negative results.

168. Report invalid runs.

169. Keep zeroes.

170. No prompt hinting unless branded/comparison test requires.

171. No domain filters in organic benchmark.

172. Don't repeat until brand appears.

173. Don't change model only when output unfavorable.

Anti-cherry-picking is foundational.

Bagian W: Reporting

174. Start with executive summary.

175. Show core metrics.

176. Show methodology.

177. Show biggest gains.

178. Show biggest risks.

179. Show source drift.

180. Show accuracy incidents.

181. Show business outcome separately.

182. Show limitations.

183. End with 3-5 actions.

Not 50.

Audit should produce decisions.

Bagian X: What Not to Claim

184. “Perplexity ranking #1” unless actual surface/metric supports.

185. “Perplexity authority score” as official.

No.

186. “Schema guarantees citation.”

No.

187. “Every citation is endorsement.”

No.

188. “Citation means claim is true.”

No.

189. “Our panel equals market share.”

No.

190. “More citations guarantee more revenue.”

No.

191. “Old source means Perplexity stale.”

Not necessarily.

192. “Third-party source is always worse than owned.”

No.

Precision protects credibility.

Bagian Y: Final Scoring

If organization wants score, use categories.

Search Presence.

Source Quality.

Accuracy.

Reputation.

Freshness.

Business Impact.

Technical/Data Readiness.

Each 0-5 internal.

But label:

Internal Audit Score.

Not Perplexity metric.

Provide raw observations underneath.

A score should summarize.

Never replace evidence.

Bagian Z: Final Decision

At end audit, classify:

PASS:
brand/source representation broadly accurate and useful.

PASS WITH ACTIONS:
some gaps, no critical risk.

RISK:
important wrong facts or source issues.

INSUFFICIENT DATA:
panel/sample not enough.

Then actions.

Example:

RISK.

Reason:
Perplexity repeatedly reports old CEO and uses two outdated directories.

Action:
correct directories, strengthen current leadership page, feedback, retest in 14 days.

That's useful.

Not:

“GEO score 63.”

What does a mature Perplexity audit feel like?

Less exciting than screenshot deck.

More like research.

Definitions.

Samples.

Sources.

Limitations.

Claims.

Actions.

That's exactly the point.

AI visibility is moving from novelty to operations.

Brand and publisher teams need methodology that survives:

source drift,

model changes,

query changes,

new data,

and executive scrutiny.

Perplexity makes verification easier by exposing citations.

The responsibility shifts to us.

Open them.

Classify them.

Check them.

Measure them honestly.

Then improve the public information ecosystem where the evidence says improvement is needed.

Audit selesai bukan ketika brand muncul.

Audit selesai ketika kita tahu kenapa visibility itu penting, apakah informasinya benar, siapa sumbernya, dan apa yang harus dilakukan berikutnya.