Perplexity prompting and source control for marketers

#01What are the four levers?

The source selector labels have changed several times. Expect at least Web, Academic, Social and Finance, plus your connected apps on paid plans.

#02How do you write source rules in a prompt?

Use a short, explicit block at the end of the prompt:

Source rules:
- Prefer: vendor documentation, pricing pages, government sites, peer-reviewed or analyst research, G2/Capterra review text.
- Avoid: affiliate "best X" listicles, AI-generated content farms, pages without a date.
- Date window: published or updated after 1 May 2026; flag anything older.
- For each claim, cite the source inline and include the publication date.

Why it works: Perplexity rewrites your question into sub-queries. Rules like "pricing page" and "documentation" change those sub-queries, so different pages are retrieved.

#03What prompt patterns work best for marketers?

1. The table-first pattern. "Return a table with columns: vendor, claim, exact quote, source URL, date. Then a 150-word summary." Tables make missing evidence obvious.

2. The named-site pattern. "Using only competitor-a.com, competitor-b.com and their help centres, list every integration they advertise." This steers retrieval strongly. Check the sources list to confirm.

3. The adversarial pattern. "What evidence contradicts this claim? Search reviews, community threads and news." This finds weaknesses in your own positioning.

4. The source-audit pattern. "List the sources you used, grouped by type (vendor, review site, community, news, government). Which useful source types did you not find?"

5. The freshness pattern. "Only include information from the last 90 days. If nothing recent exists, say so rather than using older sources."

6. The disambiguation pattern. "Ledgerline here means the accounts-payable software company at ledgerline.example, not the accounting term." Brand names that are also ordinary words need this.

#04How do files and connectors change results?

When you upload files or connect apps, Perplexity can search them alongside the web. That's powerful and risky:

• Powerful: "Compare our win-loss notes (attached) with what reviewers say publicly about Competitor B."
• Risky: Answers can mix your internal claims with public facts. Ask Perplexity to label each claim as "internal file" or "web".

Use Projects for files you reuse (see lesson 8). Projects allow up to 10,000 files and 5 GB in total, with 100 MB per file. Check what your plan allows.

#05How do API filters give strict control?

When results must be reproducible, use the API (details in lesson 11). Perplexity's web_search tool supports:

An example denylist for a market scan:

"search_domain_filter": ["-pinterest.com", "-quora.com", "-affiliate-roundups.example"]

#06What should go into Project instructions?

Instructions apply to every thread in a Project and can be up to 8,000 characters. A good marketing research instruction block:

You are a research assistant for Ledgerline's marketing team.
Company: Ledgerline, accounts-payable automation for mid-market finance teams (ledgerline.example).
Competitors: Competitor A, B, C, D (domains listed in the attached file).
Source rules: [as above]
Output: tables first; cite every claim; include dates; end with "Claims to verify".
Never invent statistics. If a number has no source, say "no source found".
UK English.

#07Worked example: Ledgerline

Ledgerline is a fictional accounts-payable software company. Details are illustrative.

Ledgerline's SEO lead asked "What do mid-market CFOs care about when choosing AP software?" The first answer cited three affiliate listicles and a 2023 blog post.

Walkthrough

1. Ran the plain question and saved the sources list.
2. Added the source-rules block and reran.
3. Added Social sources for community voice.
4. Ran inside the marketing Project with call notes attached.
5. Asked the source-audit question to see which source types were missing (analyst research was thin), then searched for it directly.

#08Common mistakes

• Rules buried mid-prompt. Put them in a clearly labelled block at the end.
• Too many named sites. Over-restricting can starve the answer. Start broad and then narrow.
• No date window. Pricing and feature research goes stale fast.
• Trusting "site:" in the app. Check the sources panel, or use the API for strict filters.

#09Try it yourself

Take one research question. Run it four ways: plain, with source rules, with a different source selector, and inside a Project with one relevant file. Score each run on the share of sources you'd trust and the share of sources dated within 12 months. Save the best prompt as a template.

#10Quick quiz

1. How many domains can `search_domain_filter` take?

Answer: Up to 20.

2. Can you mix allowlist and denylist entries in one domain filter?

Answer: No. Use one mode per request.

3. Where should rules you use on every thread go?

Answer: In Project instructions.

#11What should you check before the next lesson?

• You have a source-rules block saved.
• Your Project instructions name your competitors and source rules.
• You know when to switch to the API for strict filters.

Previous: Search, Pro Search, Research and Model Council. Next: Competitor research with Perplexity. Or return to the Perplexity for Marketers track. Related: Projects for teams, The Agent API and Search API.

#12Common questions

Can I limit Perplexity to certain websites? In the app, you can name sites in the prompt and choose source types. For strict control, the API supports a domain filter of up to 20 domains as an allowlist or a denylist.

How do I stop Perplexity using affiliate listicles? Tell it which source types to prefer and which to avoid, put the rule in your Project instructions, and rerun if the sources list still shows them.

Does Perplexity follow site colon operators? Mentioning a domain or using site-style phrasing usually steers retrieval, but it isn't guaranteed in the app. Check the sources panel or use the API filter.