Using Perplexity for Competitive Research

TL;DR: Perplexity competitive research works when you treat Perplexity as a fast way to find a competitor's public pages, not as the source of truth. Scope the questions, point searches at primary sources, verify every claim on the cited page, log it with a date, and schedule a refresh. Perplexity's own docs say its answers go wrong most on exactly the kind of information competitors publish.

What is Perplexity actually good for in competitive research?

Finding things fast. Perplexity searches, reads and cites in one step, so it can turn "what changed in Competitor X's pricing this year?" into a list of candidate pages in seconds. That is real value. Competitive research is mostly a hunt for the right pages: pricing, docs, changelogs, job listings, press releases, filings.

It is weaker at the step that matters most, which is stating the fact correctly. Perplexity's own Agent API prompt guide is candid about why: "Hallucination is most likely when the information isn't web-accessible (posts that require authentication, private documents, paywalled content), when repeated searches return related but non-matching results, or when very recent information isn't indexed yet."

Read that list with a competitor in mind. Product features often sit behind a login. Analyst reports are paywalled. Competitors with similar names produce near-miss results. And the change you care about most, last week's price change, is the most likely to be missing from the index. So the workflow below is built around one rule: Perplexity finds the page, you confirm the fact.

Why do AI tools get competitor facts wrong?

Three patterns come up again and again, and Perplexity's guide names most of them.

Near-misses presented as answers. Ask for a competitor's current plan limits and you may get last year's, or a sister product's, or a reseller's page. Perplexity suggests an instruction to "state the mismatch explicitly before answering." It is one of the most useful lines you can add to a competitive research prompt.

Pages the search cannot read. Many pricing pages are rendered in the browser with JavaScript, and many sites challenge automated visitors. We hit both while researching this cohort of posts on September 16, 2026: one vendor's site returned a bot challenge to every scripted request, and another returned a server error. An AI search tool faces the same walls, and may fill the gap from older copies or third-party summaries.

Citations that look better than they are. A citation shows where the model looked. It does not prove the sentence says what the answer claims. The largest independent test of this, which we unpack in How Accurate Are Perplexity's Citations?, found Perplexity wrong on a substantial share of one citation task, and other AI search tools did worse. The lesson is not "avoid Perplexity". It is "never skip the check".

How do you run competitive research in Perplexity, step by step?

Five stages. Each one has a concrete output, so you can tell when it is done.

Stage 1: Scope the questions (output: a question list)

Write down the decisions the research supports, then the questions that feed them. "Understand Competitor X" is not a question. These are:

• What does each plan cost, and what are its limits, today?
• Which features did they ship in the last 90 days?
• Which roles are they hiring for, and what does that suggest about their roadmap?
• Which customer segments do their own case studies target?

Keep each question narrow and give it a time range. Perplexity's guide says specific input produces more targeted retrieval, and a narrow question is also much easier to verify.

Stage 2: Set up a Project (output: standing instructions and priority domains)

For research you will repeat, use a Perplexity Project. Its settings let you "Add and manage web links and domains to be prioritized" and add instructions that apply to every query in the Project. Put the competitors' own domains in the priority list. Put your citation and grounding rules in the instructions. Our guide to Perplexity Projects covers the setup, including the Brain memory feature and sharing limits.

If you work through the Agent API instead, the same idea maps to web_search filters. Perplexity's domain filter docs allow up to 20 domains per request, in allowlist or denylist mode but not both at once.

Stage 3: Run source-first searches (output: a list of cited pages)

Ask for pages before conclusions. The first pass should return where the facts live, not the facts themselves. That keeps you from anchoring on a confident summary before you have seen the evidence.

Find the official pages where [Competitor] publishes:
1. current plan prices and limits
2. release notes or a changelog
3. open job listings
4. customer case studies

Use [competitor-domain.com] first. For each item, give the page title, the
domain, and the date shown on the page if there is one. Do not summarize
the content yet. If you cannot find an official page for an item, say
"not found" rather than using a third-party page.

For a ready library of competitor prompts to run at this stage, see the competitor section of Perplexity Prompt Templates for Research and 20 AI Prompts for Competitive Analysis. This post is about the workflow around them.

Stage 4: Verify claim by claim (output: a dated evidence log)

This is the stage people skip, and it is the one that makes the research usable. Once you have the pages, extract claims in a structured form, then check each one yourself.

From the pages you cited, list every factual claim about [Competitor]'s
[pricing / features / hiring] as rows:

claim | source domain | exact sentence from the page | date on page

Rules:
- Copy the sentence exactly. If you cannot quote a sentence that supports
  the claim, mark the row UNSUPPORTED.
- If a page is about a different year, product or company than asked,
  mark the row MISMATCH and say what it is about.
- Do not add claims that are not on the cited pages.

Then do the human part. Open each cited page. Use your browser's find function to locate the quoted sentence. If it is there and says what the row says, mark the row verified and note today's date. If it is not there, the row is out.

Your evidence log should end up looking like this:

Only verified rows go into the brief. That rule feels slow the first time. It saves you from presenting a competitor's old price to your leadership team.

Stage 5: Write the brief and schedule the refresh (output: a dated brief and a cadence)

Write the brief from verified rows only, and put the verification date at the top. Competitive facts expire. A brief without a date invites someone to act on stale numbers six months later.

For refreshes, Perplexity documents Scheduled Tasks in Computer, which the help center says "replace the older Perplexity Tasks experience". Its examples of recurring work explicitly include competitive tracking. Two limits from the same article: schedules can run no more often than once per hour, and each run uses Computer credits. The article says Max and Enterprise Max plans include monthly complimentary credits for these tasks; check what your own plan includes.

A scheduled task can flag that something changed. It should not update your brief on its own. When it reports a change, run stage 4 again on that one item.

What do you do when two sources disagree?

Sooner or later you will verify a claim and find two official pages that say different things. This happens more than people expect. Vendors update one page and forget another, and help articles lag behind product changes. Perplexity's own help center is an example: its Research article gives two different time estimates in one paragraph, and its Focus-mode article still carries that name while describing a "Choose sources" picker instead.

When a competitor contradicts itself, do not pick the answer you prefer. Handle it in four steps:

1. Log both rows. Record each claim with its own source and date, and mark both as CONFLICT.
2. Rank the sources. A live pricing or checkout page usually beats a help article, and a help article usually beats a blog post. A dated changelog entry beats an undated page.
3. Look for dates on the page. "Last modified" lines, release dates and copyright years help you tell which page is newer. A stale footer year alone proves little, so treat it as a hint, not evidence.
4. Report the conflict in the brief. "Their pricing page says X, their docs say Y, checked on this date" is more useful to a decision-maker than a confident single number that might be wrong.

A prompt can help you find conflicts, but it cannot settle them:

Compare these two pages from [Competitor]: [URL 1] and [URL 2].
List every place where they state different facts about [topic].
For each difference, quote the exact sentence from each page and give
any date shown on either page. Do not decide which one is correct.

What does a worked verification look like?

Here is the stage 4 check applied to one claim, start to finish. Say your first search returns an answer that a competitor's entry plan "includes 3 seats and costs $12 per month", citing the competitor's pricing page.

Step 1: Open the cited page. Not a cached copy or a summary. The live page, in your own browser.

Step 2: Find the sentence. Search the page for "seat" and for "12". Suppose you find "$12 per user per month" and "minimum 3 users". That is not the same claim. The answer turned a per-user price and a minimum into a flat price with three seats included, which is a real difference: the true entry cost is closer to $36 a month.

Step 3: Check the billing basis. Many pricing pages default to annual billing and show a monthly equivalent. Toggle the billing switch and note which figure you are recording.

Step 4: Log it. One row, with the corrected claim, the exact wording, the page, the billing basis and today's date. Mark the original AI claim as UNSUPPORTED in its own row, so the team can see what the first answer got wrong.

Step 5: Decide if it matters. If the brief compares entry costs, this correction changes the conclusion. That is exactly the kind of error that slips into a slide when nobody opens the page.

These numbers are illustrative, not taken from any real vendor. The pattern is real, though: per-user versus flat pricing, minimum seat counts and annual-versus-monthly display are among the most common ways a summarized price differs from the page.

When should you use Research mode instead of Search?

Use Research when you need breadth fast, such as a first map of a market you do not know. Perplexity says Research "performs dozens of searches, reads hundreds of sources" and writes a report. Its help article also says you cannot manually choose a specific model when using Research, and it gives two different time estimates in one paragraph, under 3 minutes for most tasks and around 4 to 5 minutes.

Use regular Search when you are checking specific facts. A long report creates many claims at once, and every one of them needs stage 4. Ten narrow searches are often faster to verify than one broad report. For more on picking between modes, see How to Prompt Perplexity.

What should you never use Perplexity for in competitive research?

Some jobs belong elsewhere:

• Anything behind a login. If a fact is only visible inside a competitor's product, Perplexity cannot see it, and asking may produce a guess. Use a legitimate trial account and your own eyes, within the competitor's terms.
• Paywalled analyst data. Buy the report or leave the number out.
• Private or confidential information. No tool makes that appropriate to collect.
• Final numbers in a board deck. Use Perplexity to find them. Take the number from the page itself.

What should a finished competitive brief contain?

Keep it short enough that someone will read it, and structured enough that someone can check it. A one-page brief per competitor works well:

1. Verified on: the date of your last stage 4 pass, at the very top.
2. What changed since last time: three to five bullets, each linked to its evidence row.
3. Pricing and packaging: the verified plan table, with the billing basis stated.
4. Recent launches: from the changelog, with dates.
5. Signals: hiring, case studies and partnerships, labelled as interpretation rather than fact.
6. Open questions and conflicts: anything unsupported or contradictory, stated plainly.

Keep the evidence log as an appendix or a linked sheet. When someone asks "where did this number come from?", the answer should take one click. That is the real payoff of doing competitive research this way: the brief stays useful after the person who wrote it has moved on.

How do you keep the workflow consistent across a team?

The workflow only works if everyone runs the same prompts with the same rules. Three things help:

1. One set of standing instructions, stored in the shared Project or in a shared prompt library, not copied into personal notes.
2. Templates with clear variables for the competitor name, domain and time range, so nobody edits the rules by accident when they change the target.
3. One evidence log format so verified rows from different people can be merged.

This is where a prompt library helps. Prompt Architects stores research templates with variables and can tailor a prompt for Perplexity as the target platform. The extension works on perplexity.ai, according to our integrations page, and the Team plan shares a library across 2 to 20 members. It does not search, read sources or verify anything. That stays with Perplexity and with you. If you are comparing prompt tools for this kind of work, our cohort roundup of the best prompt tools for Perplexity research workflows covers the native and third-party options.