Perplexity is an AI answer engine that searches the web and writes cited answers. It is useful for research and increasingly important as a place where brands are recommended or ignored. Its main risks for a business are inaccurate synthesis, overreliance on whatever sources it found, leaking confidential information through prompts and uploads, and misleading measurement when teams track it like a search ranking. A short governance policy agreed before wide use handles most of these.
How does Perplexity work, and why does that create risk?
For each question, Perplexity runs searches, reads a set of pages and writes an answer with numbered citations. Pro Search and Deep Research modes read more sources and run more steps. Users can upload files, create Spaces with shared instructions, and choose between several underlying models depending on the plan.
The design reduces some risks of chat assistants, because sources are shown and the information is current. It introduces others. The answer is only as good as the pages found, and the model can combine sources incorrectly or state something more confidently than the source does.
What are the main risks when teams use Perplexity for research?
Inaccurate synthesis
Perplexity can attribute a claim to a source that does not quite say it, merge two statistics from different years, or treat an opinion piece as fact. The citations make checking possible, but only if someone checks.
Source quality
The engine draws from what it finds. On niche topics, that may be a forum post, a vendor’s sales page or an outdated article. Research for client work needs primary sources where they exist: official documentation, regulators, original studies.
Confidential information
Prompts and uploaded files go to Perplexity’s servers. Pasting a client’s unpublished strategy, financial data or customer list into a consumer account may break confidentiality agreements and data protection rules. Enterprise plans offer stronger data controls, including settings that exclude data from model training; check the current terms before use.
Copyright and publisher concerns
Several publishers have challenged how AI answer engines use their content. Using Perplexity for internal research is one thing; republishing its summaries of paywalled articles is another.
What are the limits for GEO and visibility work?
Generative engine optimisation (GEO) treats Perplexity as a channel where brands can be cited. That work has its own limits:
- Answers vary between runs, so a brand may be cited one day and not the next.
- There is no official brand dashboard reporting citations or impressions.
- Results differ by language and location; English answers often draw on more sources than Turkish ones.
- Perplexity has run advertising experiments, and sponsored elements may appear alongside answers, which can blur organic and paid visibility.
- Crawler access depends on robots.txt settings for PerplexityBot and on whether pages render without heavy scripts.
Governance rules for research use
These rules suit an agency or in-house marketing team:
- Account type: client work happens only in team or enterprise accounts with training on data turned off where the plan allows.
- No confidential inputs: no unpublished client data, personal data or contract documents in prompts or uploads unless the plan’s terms and the client agreement permit it.
- Verification: every fact used in a deliverable is checked against the original cited source, and that source is cited in the deliverable instead of Perplexity.
- Source standards: prefer primary sources; flag any claim supported only by forums or vendor marketing.
- Spaces hygiene: separate Spaces per client, with access limited to the team on that account.
- Retention: delete threads containing sensitive context when the project ends.
Governance rules for GEO reporting
- Report citation share from a stated sample of queries and runs, never a single check.
- Avoid ranking language; describe “cited in X of Y sampled answers”.
- Record date, mode, model choice and location for each check, because each changes results.
- Separate sponsored placements from organic citations.
- Keep a log of inaccurate descriptions of the brand and the fixes made.
- Review robots.txt decisions for AI crawlers with the client, documenting the reasoning.
How Alien Road applies Perplexity governance
In a typical engagement the governance rules above are shared with the client during onboarding, and adjusted to the client’s own data policy. If the client prohibits external AI tools for certain data, that restriction is written into the project brief so every team member sees it.
For research tasks, strategists use team accounts, verify every claim against the original source and cite those sources in deliverables. For GEO work, the team builds a fixed question set with the client, runs it on a monthly schedule with logged settings, and reports citation share by topic with sample sizes. Inaccurate brand descriptions found during sampling are traced to their likely source, such as an old directory listing or outdated product page, and corrected at the source.
Each quarter, the policy is reviewed against changes in Perplexity’s features, terms and advertising formats, and any update is recorded.
How do you train a team to follow the policy?
Rules on a page are easy to ignore. A 30-minute session with real examples works better: show a Perplexity answer that misquotes its source, show a prompt that would leak client data, and walk through the verification step on a live question. New team members get the same session in their first week. A short quiz or a checked first deliverable confirms the habit has stuck.
Who should own Perplexity governance?
Ownership works best when split. A marketing or SEO lead owns the GEO measurement rules and the query set. A data protection or legal contact owns the rules on inputs and account types. Team leads make sure people follow both. One named person should keep the policy document current; when nobody owns it, it goes out of date within months.
Where to start this week
Check which Perplexity accounts your team uses and on which plans. Read the current data and training settings for each, and switch off training on data where possible. Then write the six research rules above on one page and share it with everyone who uses the tool. It takes an hour and prevents the most common problems before they start.
Keep reading: GEO – AIO – AEO · GEO