How to Use ChatGPT Deep Research for a Competitor Brief

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for How to Use ChatGPT Deep Research for a Competitor Brief.
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for How to Use ChatGPT Deep Research for a Competitor Brief.

To use ChatGPT deep research for a competitor brief, pick Deep research from the + tools menu (or type /Deepresearch) and paste a written brief that names three to five competitors and the exact questions you need answered. Edit the research plan, let it run, then open every cited source behind the claims you will act on.

The Complete Excel Course with AI Integration

From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.

By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner

Buy now

The brief does most of the work. A loose prompt such as "research my competitors" comes back as a long market overview with dozens of links. A brief that asks for list prices with the date seen, stated lead times and warranty terms comes back as tables you can check line by line. Checking matters because a citation only proves a page exists; it does not prove the page says what the report claims, that it is current, or that it is about the right company. Deep research is limited on the Free and Go plans, expanded on Plus at $20 a month and at its maximum on Pro, and OpenAI does not publish the number of runs each plan gets, so plan one careful run rather than five loose ones.

Every figure below comes from one illustrative business: a 14-person manufacturer of stainless-steel catering furniture (worktables, wall shelving, sinks) that sells to restaurants and caterers, partly through distributors and partly direct from its website. The owner wants one decision out of the brief: whether to raise list prices on its three best-selling worktables before the spring catalogue goes to print. Three rivals matter, called Rival A, Rival B and Rival C here.

If you want the general method for researching competitors with any AI tool, the one-afternoon competitor research method covers it. Deep research changes three things about that method: you can edit its plan before it starts, every claim arrives with a source you can open, and the whole run can be repeated from a saved project.

How do you write a competitor brief for ChatGPT deep research?

A competitor brief for ChatGPT deep research has five parts: the decision it feeds, your own position, the competitors with their web addresses, the questions, and the rules for sources. Deep research follows instructions closely, so the brief is where the quality of the report is decided. The decision line is the one most people skip, and it is the line that keeps the report short. Tell it the brief feeds a pricing decision and it stops padding the report with market-size estimates you will never use.

COMPETITOR BRIEF
Decision this brief feeds: [one sentence, e.g. "whether to change list prices on X before (date)"]
Our business: [3-4 lines: what we sell, to whom, price position, sales channels]
Competitors (research each one separately):
1. [Name] - [website]
2. [Name] - [website]
3. [Name] - [website]
Questions, answered for every competitor:
1. Product or service range in [category]: models, sizes, options
2. Published list prices: exact figure, whether tax and delivery are included, date seen
3. Stated lead times or turnaround, and minimum order quantities
4. Warranty, returns and delivery terms
5. Sales channels: direct, distributors, marketplaces
6. Standards or certifications they claim, with the page that claims them
7. Review themes: the three most common complaints and compliments,
   with how many reviews each theme rests on
8. Changes in the last 12 months: new products, price changes, new sites, closures
Source rules:
- Public sources only; give the publication or "last updated" date for each.
- Cite every factual claim. If you cannot find something, write "not found". Do not estimate.
- Label each claim "stated by the competitor" or "reported by a third party".
- Do not use my connected apps. Use uploaded files only where I name them.
Output:
- One comparison table per question, competitors as columns
- A list of claims you could not verify
- No recommendations unless I ask for them

Four of those rules do more than they look. "Date seen" exposes stale prices, because a price without a date cannot be compared with anything. "Not found, do not estimate" stops the model filling a gap with a plausible figure. The "stated by the competitor" label separates marketing claims from independent reports, so "fastest delivery in the trade" never sits in your brief as a fact. "No recommendations" keeps the report to evidence; you make the call afterwards, with checked numbers.

The same template, filled in by the catering-furniture maker

COMPETITOR BRIEF
Decision this brief feeds: whether to raise list prices on our 1200, 1500 and
1800 mm stainless worktables before the spring catalogue goes to print in six weeks.
Our business: 14-person manufacturer of stainless-steel catering furniture.
Worktables, wall shelving and sinks for restaurants and caterers. About 60% of
sales through four distributors, 40% direct from our website. Mid-priced;
10-working-day standard lead time; 2-year warranty.
Competitors:
1. Rival A - [website]
2. Rival B - [website]
3. Rival C - [website]
Questions: as in the template, with category = freestanding stainless
worktables, 1000-2000 mm wide.
Extra question: do they publish trade or distributor discounts? If yes, quote the page.
Files: our current price list (attached) is for comparison only. Never cite it
as a source for a competitor fact.

Two lines in that version earned their place in the first run. The size range stopped the report wandering into wall shelving, which the owner had not asked about. The note on the attached price list mattered because uploaded files are one of deep research's default sources; without the note, one draft table quoted the owner's own 10-day lead time against Rival C by mistake.

Adapting the questions for a service business

The template carries over to service firms with a few swaps. A courier firm comparing three regional rivals would replace "product range" with "service levels offered (same day, next day, timed)", turn question 2 into "published rates for a 5 kg parcel and a 30 kg parcel", and turn lead times into "collection cut-off times and weekend cover". A translation agency would ask for language pairs, per-word or per-page rates where published, rush surcharges, and which standards each agency says it is certified to, with the page making the claim. The source rules and the output section stay the same in every version; those are what make the report checkable.

Start the run and edit the research plan

  1. Open a new chat, or a chat inside your competitor project (set up further down), and choose Deep research from the + tools menu in the message box. Typing /Deepresearch, or choosing Deep research from the sidebar, does the same.
  2. Open Sites, then Manage sites, and enter the competitors' web addresses as a comma-separated list. You get two choices: keep the research to those domains only, or favour them while still allowing the wider web. For a competitor brief the second is usually right, because review sites and trade press live elsewhere.
  3. Attach any file you want compared, such as your own price list, and make sure the brief says how it may be used.
  4. Paste the brief and send it. ChatGPT replies with a research plan before any research starts.
  5. Edit the plan (the example below shows how), approve it, and let it run. If the activity shows it wandering, you can interrupt mid-run to narrow the focus or change which sources it uses.

OpenAI gives no fixed run time; deep research reads far more sources than a normal answer does, so start it before a meeting rather than while a customer is waiting on the phone.

A proposed plan, before and after editing

Here is the kind of plan the catering-furniture maker's brief produced, shortened and illustrative:

1. Research the commercial catering equipment market: size, growth, key trends.
2. Identify the leading manufacturers of stainless worktables.
3. Review the websites of Rival A, Rival B and Rival C for products and pricing.
4. Gather customer reviews from review platforms and forums.
5. Summarise strengths, weaknesses and strategic recommendations.

And the edited version the owner sent back:

1. Delete: market size is not needed for this decision.
2. Delete: the competitors are fixed at A, B and C.
3. For each of A, B and C: product pages for freestanding stainless worktables
   1000-2000 mm, price pages AND downloadable price-list PDFs (note the date
   printed on each PDF), plus lead time, minimum order, warranty, delivery and
   returns pages.
4. Reviews: only reviews from the last 24 months; record how many reviews
   each theme rests on.
5. Replace "strategic recommendations" with a list of claims you could not verify.

Every deletion shortens the report you will have to check, and the PDF instruction turned out to matter: price-list PDFs are where old prices hide, and asking for the printed date made the stale ones visible.

One default deserves attention before you approve. If you have apps such as Google Drive or Gmail connected, deep research can use eligible connected apps automatically when your permissions and workspace settings allow, using read actions only. For a competitor brief that is rarely what you want: a line from an internal email about a lost deal could land in a report you later forward to a distributor. The source rule in the brief covers it, and the sources-used list will show you if it happened anyway.

How do you read a finished deep research report?

Read a finished deep research report in three steps: the sources-used section first, the activity history next, then the tables, not top to bottom. A finished run opens full screen with a table of contents, a sources-used section and an activity history showing how the research progressed.

  • Sources used first. Count the sources by type. The illustrative run cited 43: 17 competitor pages, 9 review pages, 8 directory or marketplace listings, 6 trade press articles and 3 PDFs. If competitor pages make up less than about a third, the report is leaning on other people's summaries of your rivals, and summaries go stale faster.
  • Activity history next. Look for pages it tried and could not read: blocked sites, paywalled articles, PDFs it could not open. A gap there explains a "not found" in the tables and tells you where to look yourself.
  • Then the tables. Highlight every decision claim, meaning any figure or term that would change the decision. For a pricing decision that is list prices, lead times, minimum orders and warranty terms. In the worked run, 12 of the report's sixty-odd claims were decision claims. The rest (founding years, office photos, social follower counts) can stay unchecked because nothing hangs on them.

Download the report straight away in the format you will mark up. Deep research reports download as Word, PDF or Markdown; Word is the easiest to annotate with a verdict next to each claim.

How do you check a deep research claim against its source?

A deep research claim passes the citation check only when its cited page states the claim, covers the same company and product, is current, and is the right kind of source. Check one claim at a time: open the cited page, find the sentence, and ask whether the page actually says this, whether it is about the same company and the same product, whether it is current, and whether it is the right kind of source for this claim (a competitor's own price page for a price, not a directory listing). If you want a fuller method for checking AI research in general, the tutorial on checking sources and citations in AI research goes through it; the log below is the competitor-brief version.

#Claim in the report (illustrative)Cited sourceResultFix
1Rival A 1500 mm worktable lists at $689Rival A product pageConfirmed: $689, delivery extraNote "delivery extra"
2Rival A standard lead time 5 working daysRival A delivery FAQConfirmedNone
3Rival B 1500 mm worktable lists at $612Price-list PDFStale: PDF printed two years ago; site now says "price on request"Record "no current public price"
4Rival B gives a 10-year warrantyBusiness directory listingWrong company: a similarly named cold-room shelving firmRemoved; Rival B's own terms say 3 years
5Rival C trade minimum order is 2 unitsRival C trade pageConfirmedNone
6Rival C lead time 15 working daysTrade press articleStale: article 20 months old; site now says 10-12 daysUse the site figure, with date seen
7Rival C 1500 mm worktable lists at $655Rival C product pageConfirmed, includes deliveryNote "delivery included"
8Rival A offers 12% trade discountRival A trade sign-up pageUnsupported: page invites trade sign-up, no percentage shownRecord "not published"

Across all 12 decision claims the illustrative tally was 8 confirmed, 2 stale, 1 wrong company and 1 unsupported. Checking took about 50 minutes, a little over four minutes a claim, most of it spent finding the right line on long pages. That is the honest cost of a cited report: the citations make checking possible, but they do not do it for you.

Notice how much the delivery notes change the picture. Before checking, Rival C looked $34 cheaper than Rival A. After checking, Rival C's $655 includes delivery and Rival A's $689 does not, so the real gap is wider than the headline figures suggest. A report that compares list prices without that detail would have pushed the owner towards the wrong comparison.

How the wrong-company citation gave itself away

The 10-year warranty claim for Rival B looked plausible and came with a tidy citation, which is exactly why it nearly survived. The giveaway was on the cited page itself: the directory entry listed cold-room shelving and a different address and phone number from the ones on Rival B's own contact page. Deep research had matched on a near-identical trading name. Two habits catch this quickly: compare the cited page's contact details with the competitor's own site, and list each competitor's domain under Manage sites so their own pages outrank lookalikes.

Fixing failed claims without a fresh run

Do not rerun the whole brief to fix four claims. Ask a targeted question in a normal chat with web search, or as a follow-up in the same conversation:

Claims 3, 4, 6 and 8 failed checking:
3 - the price-list PDF is two years old; Rival B's site now says "price on request".
4 - the warranty source is a different company.
6 - the lead-time article is 20 months old.
8 - the cited page shows no discount percentage.
Re-check only these four against each competitor's own website. For each,
give the exact wording, the page address and the date you saw it.
If the competitor does not publish the figure, say "not published".

That kind of follow-up costs a fraction of the time and leaves your deep research allowance for the next full brief.

If you keep the claim check log in a spreadsheet and want it quicker to build, try The Complete Excel Course with AI Integration on Coding Liquids.

How do you turn a checked report into a one-page competitor brief?

Turn a checked deep research report into a one-page brief by attaching the marked-up check log and asking for a summary built only from confirmed claims. The raw report is evidence; the one-pager is what people read. Use a prompt like this:

Using ONLY the claims marked Confirmed or corrected in the attached check log,
write a one-page competitor brief for this decision: whether to raise list prices
on our 1200, 1500 and 1800 mm worktables.
Structure:
1. Price table: our price and each rival's, with "delivery included/extra"
   and the date seen. Show "no current public price" where that applies.
2. Lead times and minimum orders, same format.
3. Warranty terms.
4. Three review themes per rival, each with the number of reviews behind it.
5. What we could not verify.
Keep every figure exactly as in the log. Do not add recommendations.

An illustrative extract of what came back:

1500 mm worktable, list prices (seen this week): Ours $629, delivery extra. Rival A $689, delivery extra. Rival C $655, delivery included. Rival B: no current public price.

Lead times: Ours 10 working days. Rival A 5. Rival C 10-12. Rival B not published.

Summary: We are the lowest-priced option in the market and could raise prices by around 8% without losing our position.

Two things to fix before this goes anywhere. The summary line breaks two rules at once: "lowest-priced in the market" ignores that Rival B's price is unknown, and the 8% suggestion is a recommendation the prompt banned. Delete both and keep the tables. With checked numbers in front of them, the owner (in this illustration) raised the 1500 mm table by 5% to $660, still below Rival A's $689 before delivery and close to Rival C's delivered $655, and kept the 10-day lead time as the point to improve, since Rival A ships in 5.

How do you save a competitor brief as a ChatGPT Project?

Save a competitor brief as a ChatGPT Project, which keeps the template, the rules and the previous brief in one place and is available on every plan, Free included. Rebuilding the brief every quarter wastes the work you have just done. If Projects are new to you, the tutorial on customising ChatGPT with Projects, instructions and files covers the basics; this is the competitor-brief setup.

  1. Select New project in the sidebar, name it (for example "Competitor briefs: worktables") and pick an icon and colour.
  2. Open the ••• menu, choose Project settings and paste the source rules from the brief as project instructions. Project instructions apply only inside that project and override your global custom instructions.
  3. In the same settings, set Memory to Project-only memory if you do not want other chats feeding into competitor work. OpenAI notes the change can take a few hours to apply.
  4. Upload four files: the blank brief template, your current price list, the checked one-page brief and the claim-check log. That uses 4 of the project's file allowance: 5 on Free, 25 on Go and Plus, 40 on Pro and Business. Only 10 files can be uploaded at once.
  5. When a project chat produces something worth keeping, open that response's message menu and choose Add to project sources, so the next run can see it.

On paid plans, deep research can run from inside the project; OpenAI's wording is that paid plans "may include" it depending on the subscription, so check that the option appears in the project's tools menu before you plan around it. Next quarter, the whole run starts with one message:

Run the competitor brief in the project files for [quarter]. Same competitors,
same questions, same source rules. Then compare with the attached previous
brief and list every change, each with its source and the date seen.

The comparison is where the project pays for itself. In the illustrative second run, the "what changed" list flagged two items: Rival A had raised its 1200 mm and 1500 mm tables by about 6%, and Rival C had launched a quick-ship range promising dispatch in 3 working days. Both went through the same four-question check before the owner believed them, and the quick-ship range changed the next brief's question list, which now asks about express options for every rival.

Sharing works on every plan, with collaborator caps set by the owner's plan: 5 on Free, 10 on Go and Plus, 100 on Pro. Shared projects are switched to project-only memory automatically. Members can see the project's chats, files and member list, so a sales manager can read the checked brief without asking for a copy. If your team then wants talking points for calls, competitor battlecards a small sales team can use are the natural next step.

How many deep research runs does each ChatGPT plan allow?

OpenAI lists ChatGPT deep research by level rather than by number: limited on Free and Go, expanded on Plus and maximum on Pro, and its help page says only that usage varies by plan. Your in-product usage counter shows the tasks you have left, and on plans with a fixed monthly allowance the counter resets every 30 days from the date you first used it. ChatGPT Business is listed with a two-seat minimum on OpenAI's pricing page, though its help centre says a new workspace may be able to start with one paid seat (both checked 20 November 2026).

PlanPriceDeep researchApps as deep research sourcesFiles per project
Free$0LimitedLimited5
Go$8 a monthLimitedLimited25
Plus$20 a monthExpandedIncluded25
ProFrom $100 a month (three usage tiers)MaximumIncluded40
Business$25 per standard seat monthly, $20 billed annually (a two-seat minimum on OpenAI's pricing page; its help centre says a new workspace may start with one paid seat)IncludedIncluded40

Prices and levels are from OpenAI's pricing pages as checked in November 2026. Three habits stretch whatever allowance you have:

  • One brief, all competitors. Three rivals in one run cost one task; three separate runs cost three and give you three reports to reconcile.
  • Edit the plan before approving. Cutting the market-size step and the recommendations step made the report shorter without losing any decision claim.
  • Use normal chat for fixes and spot checks. A single price or a single warranty term does not need deep research; web search in a normal chat finds it, and you check it the same way.

Where do deep research competitor briefs still mislead?

Deep research competitor briefs still mislead through quote-only rivals, resellers' prices, thin review evidence, old news, another currency and marketing language, and each has a check that catches it. Citations remove some risk, not all of it. These are the failure modes that showed up most often in competitor work.

  • Quote-only competitors. Many firms publish no prices. Deep research cannot see a private quote, and the right entry is "not found". If your report shows a price for a quote-only rival, it came from an old PDF, a reseller or a guess; trace it before you use it.
  • Resellers' prices passed off as list prices. A distributor's discounted or marked-up price can be cited as the manufacturer's price. Check that the cited domain belongs to the competitor.
  • Thin review evidence. "Customers complain about delivery damage" may rest on 7 reviews out of 31. The brief's rule to give review counts keeps that visible; anything under about ten reviews is an anecdote, not a pattern.
  • Old news presented as current. A launch article from two years ago can describe a product line that has since been dropped. The publication date on every source is what catches it.
  • Another currency or another market. Some competitors run separate sites for export customers with different prices. If a price looks oddly high or low, check the currency and which of the competitor's sites the citation points to.
  • Marketing language as fact. "Industry-leading lead times" is a claim, not a lead time. The "stated by the competitor" label stops it sitting in your table as a figure.

Between quarterly briefs, prices move. If your market reprices often, set up a lighter watch on the two or three pages that matter, as described in tracking competitors' prices and offers with AI, and keep deep research for the full brief.

How long does a competitor brief take and what does it cost?

A first competitor brief took the illustrative manufacturer roughly 1 hour 50 minutes of its owner's time on a Plus plan at $20 a month, and later briefs about 1 hour 20 minutes. Here is the sum for the first run:

  • Writing and filling in the brief the first time: 30 minutes. Reusing it from the project next quarter: 5 minutes.
  • Editing the research plan: 10 minutes.
  • The run itself: waiting time only; the owner did other work.
  • Checking 12 decision claims: about 50 minutes.
  • Fixing four failed claims and producing the one-page brief: 20 minutes.

That is roughly 1 hour 50 minutes of owner time for the first brief and about 1 hour 20 minutes for later ones, plus a subscription most owners already use for other work. The figure that matters is the one in the price table: a 5% rise on the best-selling worktable, set with checked competitor prices rather than guessed ones. If a run ever produces a decision claim you cannot check in the time you have, leave it out of the one-pager; an unchecked number in a pricing brief costs more than a gap.

If you want help fitting deep research into how your business actually runs, I offer one-to-one AI consultations for small businesses.

ChatGPT deep research competitor briefs: questions owners ask

Can deep research find prices that competitors only give on request?

No. Deep research reads public pages, files you upload and apps you have connected, so a price that only appears in a private quote is out of reach. The honest answer in the report is "not found", which is why the brief asks for exactly that instead of an estimate. If you need those figures, use what your sales team hears from customers, recorded with the date, rather than anything the model guesses.

Will deep research read my Gmail or Drive while it researches competitors?

Yes, it can. OpenAI says eligible connected apps are available to deep research automatically when your permissions and workspace settings allow, and it uses read actions only. For a competitor brief, add a source rule telling it not to use connected apps, then confirm in the sources-used section of the report that no internal emails or documents appear before you share it.

How do I share the finished report with my team?

Download it as Word, PDF or Markdown from the report, or run the brief inside a shared project. Project members can see the project's chats, files and member list, and sharing works on every plan, with collaborator caps of 5 on Free, 10 on Go and Plus and 100 on Pro. Share the checked one-page brief rather than the raw report, so nobody quotes an unchecked claim.

How often should I rerun a competitor brief?

Quarterly suits most small firms, though you should match it to how fast your market's prices and terms move. Time the run a few weeks before price or catalogue decisions. Between runs, a simple price watch on two or three pages catches sudden moves. Rerunning monthly mostly burns your deep research allowance on reports that barely change, and every run still needs its decision claims checked.

How do I use ChatGPT deep research for a competitor brief?

Pick Deep research from the + tools menu, then paste a written brief that names three to five competitors, the decision the brief feeds, the exact questions and the rules for sources. Edit the research plan ChatGPT proposes, let it run, then open every cited source behind the claims you will act on before the report reaches anyone.

Which ChatGPT plans include deep research?

Deep research is limited on the Free and Go plans, expanded on Plus and at its maximum on Pro, and the table above shows it included on Business. OpenAI does not publish the number of runs each plan gets, so check the usage counter in the product and plan one careful run rather than five loose ones.

Further reads

Sources: OpenAI help pages on deep research, Projects in ChatGPT and file uploads; the ChatGPT pricing page and the ChatGPT Business pricing page, all checked 17 November 2026.

Rather not work this out alone?

An AI implementation consultation is a one-to-one call with me, not an agency or a sales team. You leave knowing what AI you can use in your business and where to start.

Book a 1:1 call with me