Perplexity vs. ChatGPT for research

Editor's note: This is a documentation-led edition. We have not yet run our own hands-on tests of Perplexity or ChatGPT. What follows rests on official documentation, published pricing and independent studies we read in full, all accessed on 16 October 2026. A tested update will follow, based on our citation test.

Perplexity calls itself an answer engine: you ask, it searches, and it replies with numbered sources. ChatGPT is a general chat assistant with web search and a deep research mode built in. Both tools work in two ways. A quick search gives a short answer with links in seconds. A deep research run takes longer and returns a long report.

The documentation shows real differences: how much you can steer a research run, how sources are labelled, which plans include what, and whether your questions can train models. This comparison covers those. It does not say which tool gets more answers right. We have not measured that yet, and vendor documentation cannot tell you.

What each is for

Perplexity searches the internet in real time and returns "a clear, concise summary" with citations, its help centre says.[1] Research is its in-depth mode, pitched at complex questions in areas such as finance, health and current affairs.[2]

ChatGPT is a general assistant that can also search. OpenAI's help centre puts the split simply: "Use search for quick facts, and use deep research for depth and thoroughness."[3] Deep research is meant for multi-step questions, "especially when you want explicit control over which sources are used."[3]

Our shortlist of the best AI research tools files both under deep web research.

Sources: notes 1–19. US list prices from official pages; plans and limits change often.

Search mode: quick answers with links

ChatGPT search can start automatically when a question would benefit from current information, or you can choose Search from the tools menu. Responses "may include citations". You select one to open the source, and on desktop web you can hover over it to preview. A Sources button lists the cited pages plus "other relevant links".[4]

OpenAI also explains where your query goes. ChatGPT search sometimes rewrites your question into shorter queries and sends them to outside search providers, with a general location based on your IP address, but not the address itself or your account details.[4]

Perplexity's default search is free in practice: the free plan lists "practically unlimited basic searches".[5] Pro Search goes further. Perplexity says it conducts "thorough research across multiple sources" and, on paid plans, lets you pick which AI model writes the answer.[1] Free users get 3 Pro Searches a day.[5]

Deep research mode: long reports

Both deep modes search repeatedly and write a structured report. They differ in where you can step in.

ChatGPT deep research starts by proposing a research plan, which you can review and edit before it begins. While it runs, you can follow progress and interrupt it "to refine the focus, including adjusting which sources it can access." The finished report opens with a table of contents, a "sources used" section and an activity history, and it downloads as Markdown, Word or PDF.[3]

Perplexity Research performs "dozens of searches, reads hundreds of sources", according to its help centre, and you cannot pick the model it uses.[2] An update called Advanced Deep Research added clarifying questions for broad queries, follow-up questions while the research is still running, a progress view showing which sources are being read, and key findings that appear before the report is done. Reports now stream into a file you can edit and share.[6] Perplexity says it tested the update against industry benchmarks and that it "matches or outperforms competitors". That is a vendor claim. We have not seen the benchmark details.[6]

Perplexity says most reports take a few minutes.[1][2] OpenAI gives no typical duration.[3]

How citations are shown

Perplexity says every answer includes "numbered citations linking to the original sources".[1] Some cited domains now carry a shield icon with one of three labels: Government, Academic or Trusted. Perplexity rates whole websites, not individual pages, using questions such as whether a site corrects its mistakes. It says partnerships and payments do not affect labels, and most domains have none.[7] Pro also promises "10x as many citations per answer".[8] As the evidence below shows, more citations is not the same as better support.

"A label is never a substitute for reading the source yourself."

— Perplexity Help Center, Understanding source labels (2026).[7]

ChatGPT search shows inline citations with hover previews and a Sources panel.[4] Deep research reports collect references in a "sources used" section and keep an activity history, so you can see how the research progressed.[3] OpenAI's help page is blunt about the limits:

"Search results and citations can be incomplete, outdated, or incorrect."

— OpenAI Help Center, Searching the web with ChatGPT (2026).[4]

Source controls: what you can steer

This is where the documentation differs most.

ChatGPT deep research gives the most explicit site control of the two. Under Sites, you can restrict research to domains you enter, or choose to prioritise those sites "but allow full-web search". It can also use uploaded files and connected apps such as Google Drive or SharePoint. It uses only read actions from those apps, and the connections do not override provider permissions or workspace restrictions.[3] Ordinary ChatGPT search has no documented site filter. OpenAI suggests asking it to search again "using a specific source, date, or location."[4]

Perplexity offers a "Choose sources" control. On the help page we read, the options are web sources only or none, plus organisation-file options that are being retired.[9] Premium sources can be selected with @ or the + menu. Wiley, Statista, PitchBook, CB Insights and Midpage are included for all users, with daily limits by plan. Paid users can also connect licensed services they already subscribe to, such as FactSet or Morningstar.[10] Inside a Project, you can add web links and domains to prioritise for every query.[11]

AI Chronicle

Prices and limits (as of 16 October 2026)

Perplexity. Free includes 3 Pro Searches a day and 1 Research query a month.[5] Pro costs $20 a month or $200 a year, and Max $200 a month or $2,000 a year.[12] The pricing page shows Pro at $17 a month and Max at $167 a month when billed annually.[13] Education Pro is $10 a month with student verification. Enterprise Pro starts at $40 a seat a month and lists 50 Research queries a month. Enterprise Max is $325 a seat a month.[5][12] Pro and Max limits are described only as "average use" and "advanced use".[5]

ChatGPT. The pricing page lists deep research as "Limited" on Free and Go, "Expanded" on Plus and "Maximum" on Pro. Dollar figures did not render on the page we fetched.[14] OpenAI's help centre gives Plus at $20 a month, with no annual option.[15] Pro comes in three monthly tiers: $100, $200 and $500.[16] Deep research allowances reset every 30 days from first use on plans with a fixed monthly allowance, and availability depends on your country.[3]

Neither vendor publishes an exact number of deep research runs for its main paid plans on the pages we read.

Data use and training

ChatGPT. On Free, Plus and Pro, OpenAI "may use your content to train our models" unless you turn off Improve the model for everyone or opt out through its Privacy Portal. There is a catch: if you give a thumbs up or down, "the entire conversation associated with that feedback may be used" for training, even after you opt out. Temporary chats are not used for training but may be kept for up to 30 days for safety. Business, Enterprise and Edu content is not used for training by default.[17][18] Deep research follows the same settings as regular chats.[3]

Perplexity. For Free, Pro and Max users, AI data retention is on by default, which means your data is collected for AI training until you switch the toggle off. Opt-outs apply only from that date, and Perplexity says "previously collected training data cannot be deleted or removed." Enterprise data is never used for training, and Enterprise uploads are kept for 7 days.[19]

For both tools, opt out before you paste anything sensitive, or use a business plan your organisation has approved.

What independent studies found

Two large studies tested both tools, with limits that matter.

Tow Center, March 2025. Researchers gave eight AI search tools excerpts from news articles and asked for the headline, publisher, date and URL. Perplexity answered 37% of queries incorrectly. ChatGPT search misidentified 134 of 200 articles and signalled uncertainty only 15 times. Perplexity Pro answered more prompts correctly than the free version but also gave more confidently wrong answers. Both tools sometimes cited syndicated copies instead of the original article.[20]

EBU and BBC, October 2025. Journalists from 22 public broadcasters assessed answers to news questions from the free versions of four assistants, generated in late May and early June 2025. Significant issues of any kind affected 36% of ChatGPT's responses and 30% of Perplexity's. Significant sourcing issues affected 24% and 15% respectively. Evaluators "often felt Perplexity was strong on sourcing", but they also flagged long source lists that the answer never referred to. In one case, 19 URLs were listed and three were used. The report also gives examples of quotes Perplexity altered and quotes ChatGPT attributed to the wrong person.[21]

What these studies can't tell you. Both tested news questions, not literature reviews or market research, and neither tested a deep research mode. The Tow Center ran each query once; responses vary between runs. The EBU study used free versions and asked assistants to favour the broadcaster's own sources. Both tools have changed since. What the studies do show: in both products, a tidy citation did not reliably mean the source said what the answer claimed. Our AI hallucinations explainer covers why.

Documented limits and risks

• Usage caps are vague. Perplexity says advanced-model access "may be limited during weeks of especially heavy usage."[8]
• Less control over Perplexity Research. You cannot choose its model, and its help pages describe no way to edit a research plan or restrict sources to a site list.[2][6]
• ChatGPT search filters are informal. Site restriction is documented for deep research only.[3][4]
• Labels describe sites, not claims. A Trusted label rates a website as a whole. It does not check whether that page supports the sentence it is attached to.[7]
• Training defaults are on. Both need a manual opt-out on personal plans.[17][19]

Who it suits, and who should skip it

Perplexity suits people who want citations by default, a free taste of deep research (one Research query a month), labelled sources and built-in premium data such as Statista or Wiley.[5][7][10] Skip it if you need to lock a research run to approved sites, or if your work needs to see and edit the research plan before it runs.

ChatGPT suits people who already draft and analyse in it and want deep research they can steer: an editable plan, site restriction, connected apps and an activity log.[3] Skip it if you want Perplexity-style source labels, or a cheaper entry point to deep research: Plus is the lowest plan listed with more than "Limited" access.[14]

For confidential work, either tool's business tier excludes training by default, so let admin controls and retention settings decide.[18][19] For the wider picture of each product, see our Perplexity review and ChatGPT review.

How to test it yourself

1. Pick three questions you can already answer from trusted sources: one factual, one contested, one recent.
2. Run each in both quick modes with the same wording, then in both deep modes.
3. Score five citations per answer. Does the link open? Is it the original source, not a copy? Does the cited page contain the claim?
4. Count unused sources. How many listed links never support a sentence in the answer?
5. Test the source controls. In ChatGPT deep research, restrict to two domains. In Perplexity, use a Project with prioritised domains. Check whether the citations stay inside those domains.
6. Ask a question with no good answer. Note whether each tool says so or fills the gap.
7. Check the data settings on your account before and after, and record which plan you used.

A worked example (hypothetical)

This scenario is invented to show how the documented features fit a task. No tool was run, and no outputs are reported.

Omar, a policy officer at a transport authority, needs a two-page note on how other cities price congestion charges, citing only government and transit-agency sources.

In ChatGPT, he would start a deep research run, restrict it to the agency domains he trusts, and edit the proposed plan to drop a section on private-sector toll roads. Midway, the activity view shows a detour into news commentary, so he interrupts it and narrows the sources again.

In Perplexity, he would set up a Project with those domains prioritised and run Research. When it asks a clarifying question, he names the cities he cares about. He would watch for the Government label on citations, while remembering it rates the site, not the claim.

Either way, he then opens every citation, finds the sentence behind each figure and cuts anything he can't trace. The tool chooses where to look; he decides what goes in the note.

Questions readers commonly ask

Which is more accurate?

We cannot say yet. The 2025 studies found fewer significant issues for Perplexity than for ChatGPT on news questions,[21] but they did not test deep research or non-news tasks, and both tools have changed since. Our citation test will measure this directly.

Can I use both for free?

Yes, with limits. Perplexity's free plan includes 3 Pro Searches a day and 1 Research query a month. ChatGPT Free includes web search and "Limited" deep research.[5][14]

Does Perplexity use ChatGPT's models?

Perplexity Pro lets you choose models from several companies, including OpenAI, for Pro Search, but Research picks its own models.[2][8]

Can I stop either tool training on my research?

Yes. On personal plans, turn off "Improve the model for everyone" in ChatGPT or "AI data retention" in Perplexity. Business tiers of both exclude training by default.[17][19]

The next step

Whichever tool you pick, the report is a starting point, not finished evidence. Before it reaches a colleague or client, work through our guide on how to verify AI-generated research.

Sources and evidence note

Product descriptions, prices, limits and data terms come from Perplexity's and OpenAI's own help centres and pricing pages. They describe intended behaviour, not tested performance. Prices are US list prices as displayed on 16 October 2026. Independent evidence comes from the Tow Center (Columbia Journalism Review) article and the full EBU/BBC report, each read in full. The worked example is invented. All pages were accessed on 16 October 2026.

Related reading

• Claude vs. ChatGPT vs. Gemini: choosing an assistant
• Build a research agent with source verification