Playbooks · Everyday tasks · last verified 2026-10-23

Research and compare two big purchases

For a household buyer × comparing two specific products before buying one × with ChatGPT search, checking every price and spec at the source.

Before you start

You need the two exact products: brand, model name, and model number if there is one. "Two washing machines" is not a comparison; "Brand A Model X123 versus Brand B Model Y456" is. Write down what the purchase must do, what it must fit (dimensions, connections, space), and your budget ceiling as a number you set — the chatbot does not set it for you.

Time and cost: NOT_MEASURED. Prices change by store and date, so this page states no price for any product and no saving from using the method.

Do not paste payment details, account logins, or addresses into the chat. A product comparison needs model numbers, not personal information.

What "done" looks like

At the end of one sitting you have a side-by-side table for the two exact models, with a source link next to each spec and price, plus a short list of questions still open — and you know which of those questions only the seller or manufacturer can answer. You have not bought anything yet.

Steps

1. Name both models and ask for a sourced table. Paste:

   Compare these two exact models: [Brand A Model X123] versus
   [Brand B Model Y456]. Search the web for current manufacturer
   specifications. Make a table with one row per feature: dimensions,
   capacity, energy rating, warranty length, and what's included in
   the box. Put a source link next to every row, and write UNKNOWN
   where you cannot find a manufacturer or retailer source.

   OpenAI describes ChatGPT search as pulling the latest information from the internet into a conversation, with citation links to review in the original sources. The globe icon next to a response indicates search was used.

2. Separate specs from prices. Specs come from the manufacturer; prices come from sellers and change often. Paste:

   Now list current prices for each model from two retailers each,
   with the retailer name, the price you found, and the source link.
   If the price is a sale price, say so and note any end date shown
   on the retailer page. Do not average prices and do not guess.

3. Ask what breaks the tie for your constraints. Paste:

   Given my constraints — it must fit a space [width x depth x
   height], it must [must-have feature], and my ceiling is [amount]
   — which of the two fits, and what fails each constraint? Answer
   only from the table above and say UNKNOWN where the table has
   no answer.

4. Open the sources yourself. Click through to the manufacturer spec pages and the retailer pages. Confirm the model number matches the product you intend to buy — a single digit difference can be a different model — and confirm the price, warranty, and included items on the seller's page as of today.
5. Write the open questions down. Anything marked UNKNOWN, plus anything the sources disagree on, goes on a short list to ask the seller or manufacturer directly. A comparison with open questions is a finished first sitting; buying with them unanswered is not.

Check the agent's work

• Every row in the table names the exact model you asked about. A row sourced to a different model, a review roundup, or a generic product line does not count.
• Every price was opened on the retailer's own page by you, and matches what the table says — including whether it is a sale price.
• Dimensions, capacity, and warranty figures come from the manufacturer where possible, not from a retailer listing or a review alone.
• Any feature the table claims for both models is actually comparable: the same measurement, the same unit, the same conditions.
• The recommendation, if one is given, follows from your stated constraints — not from popularity, ratings, or a feature you did not ask for.

What can go wrong

• It compares the wrong model. Similar model names are the classic error. Checking the model number on the source page is the fix, and it is a check only you can do.
• It states an old price as current. Search results can carry stale prices. The retailer page you open today is the price; the chatbot's table is a lead to it, not a substitute.
• It fills gaps instead of marking them UNKNOWN. A confident row with no source link is the signal to distrust that row and check it yourself.
• It turns a comparison into a recommendation you did not ask for. If you only wanted facts, a recommendation adds a judgement you did not request. Ask it to stick to the table and decide yourself.

Limits

• ChatGPT search does not replace specialised sources: OpenAI's own page notes that search will not replace specialised databases. For safety certifications, recall notices, or installation requirements, go to the manufacturer or the relevant authority directly.
• The chatbot cannot buy the product, check live stock at your local store, or confirm delivery to your address in this playbook. Those are checks you do with the seller.
• Always review linked sources before making decisions — OpenAI states this directly about search citations, and this playbook follows that rule in step 4.

Verification

Checked 2026-10-23. This page was not run end to end; its capability claims rest on provider documentation fetched this run. Sources: OpenAI — Research with ChatGPT (ChatGPT search pulls the latest information from the internet into conversations; click Web Search from the tools menu; globe icon indicates search was used; click citation links to review original sources; "always review linked sources before making decisions"; search will not replace specialised databases). Omitted: no claim is made that the chatbot can purchase, check local stock, or confirm delivery — those steps were cut because they are not grounded in the fetched page, and the playbook says so in Limits. Limits of this verification: provider pages describe search behaviour, not whether any given price or spec is current — that check is the "Check the agent's work" section above.

Back to: Everyday playbooks hub · Skills index