Either works for listing copy and client emails, so the choice rarely turns on prose; for market reports it turns on file handling and arithmetic. Claude Pro and ChatGPT Plus both cost $20 a month, so pay for both for one month, run the same five tasks on each, and keep the one whose drafts you edit least.
Two caveats shape the rest. Neither assistant knows your market data unless you hand it an export, and neither should see a buyer's bank statements on a personal plan with training switched on. Most of the real difference sits in the plumbing: how many pages of a disclosure pack each one reads, whether a team can share a Project, and what each plan does with your chats.
Five tasks to run on both, in the order a realtor's day goes
Run these on one afternoon, in fresh chats, on the same tier. Paid against paid is the practical choice, because ChatGPT Free allows only three file uploads a day and this test uploads a PDF, a CSV and a dozen photos. Paste identical inputs and the identical prompt. Resist the urge to nudge the wording for one tool; the test is about the assistants, not your prompting. After each task, write down how many minutes you spend editing the draft until you would send it.
- Open-house follow-ups. Input: your shorthand notes on three visitors (what they liked, budget band, main objection). Ask for three short emails, each with one next step. Check for invented details about the visitor, and for tone: does it sound like you, or like a brochure?
- Listing description. Input: your walkthrough notes, the feature sheet and a dozen photos. Ask for 150 words using only the supplied facts. Check every adjective against your notes.
- Seller update. Input: showings this week, two pieces of buyer feedback, three comparable listings. Ask for a plain update ending in a recommendation the seller can accept or change. Check that the recommendation is framed as the seller's decision, not a pricing verdict.
- Inspection and disclosure digest. Input: a PDF of about 90 pages. Ask for defects ranked by likely cost, each with its PDF page number. Check page numbers and costs one by one.
- Market summary. Input: a CSV of 40 closed sales from your own system. Ask for median price, median days on market and a three-sentence read. Check the arithmetic in a spreadsheet.
Here is the prompt I would use as the shared input for task two. The constraint lines matter more than the tone line.
You are helping a real estate agent write a listing description.
Use only the facts below. Do not add any feature that is not listed.
Length: 140 to 160 words. Tone: warm, specific, no cliches.
Describe the home, never the people who might live in it.
End with one line listing anything you were tempted to add but could not verify.
Facts:
- 3 bed, 2 bath, 1,420 sq ft, built 1998
- Kitchen refitted 2022: quartz worktops, new oven
- South-facing garden, 40 ft, new fence 2024
- Single garage, driveway for 2 cars
- 8-minute walk to a school; bus stop on the corner
- Boiler replaced 2021; double glazing throughoutBelow is an illustrative first draft, written to show the slips you are hunting for rather than copied from a particular run.
Welcome to this sun-drenched three-bedroom home, perfect for a growing family. The granite kitchen was refitted in 2022 and opens onto a generous south-facing garden. Steps from the best schools, with a single garage and driveway parking for two cars. Move-in ready, with a recently replaced boiler and double glazing throughout.
Four sentences, four problems. "Granite" contradicts the notes, which say quartz. "Perfect for a growing family" describes who should live there, the kind of wording most likely to cause trouble under fair-housing rules. "The best schools" is a claim you cannot support, because your notes only give an eight-minute walk to a school. "Move-in ready" appeared from nowhere. The fixed version reads: "A three-bedroom, two-bathroom home of 1,420 square feet, built in 1998. The kitchen was refitted in 2022 with quartz worktops and a new oven, and the 40-foot south-facing garden has a new fence (2024). There is a single garage plus driveway space for two cars, an eight-minute walk to a school, and a bus stop on the corner. The boiler was replaced in 2021 and the windows are double glazed throughout."
Tally three things per tool: invented facts, phrases that steer by family status or neighbourhood character, and unsupported superlatives. That tally tells you more than a general feeling about better writing. The final line of the prompt, asking the assistant to list what it was tempted to add, is worth keeping even after the test. It turns hidden guesses into a visible list you can delete.
Score the drafts blind. Paste each tool's output into a document labelled A and B, with the tool names removed, and ask a colleague or a friend in the trade to mark which one they would send. People read drafts differently once they know which brand wrote them, and a blind pick is the cheapest guard against your own preference.
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 nowOne agent's month: what the five tasks save and what $20 buys
Take a solo agent with four new listings a month, about twenty buyer follow-ups a week, sixteen seller updates a month (four listings, weekly), three disclosure packs and two market summaries. The minutes below are my assumptions for illustration. Replace them with your own stopwatch numbers from the test.
| Task | Jobs a month | Minutes each before | Minutes each after editing | Minutes saved |
|---|---|---|---|---|
| Listing description | 4 | 30 | 12 | 72 |
| Buyer follow-up email | 80 | 5 | 2 | 240 |
| Seller update | 16 | 15 | 6 | 144 |
| Disclosure pack digest | 3 | 60 | 25 | 105 |
| Market summary | 2 | 75 | 30 | 90 |
| Total | 1,090 a month | 439 a month | 651 (about 10.9 hours) |
If an hour of your time is worth $35, those 651 minutes are worth about $380 against a $20 plan. The plan price is roughly five per cent of the value, so it should not drive the choice. Editing minutes should. If one assistant saves you 10 per cent fewer minutes across the month (about 65 of the 651), that gap alone is worth around $38, more than the subscription. This is why the same-afternoon test with a stopwatch beats any comparison table, including the ones in this tutorial.
The biggest single saving in the table is the follow-up email, because of volume. Short emails from simple notes leave the least room for the two to differ, so do not expect the test to split on them. If it splits anywhere, look first at the disclosure digest and the market summary, where file handling and arithmetic come into play.
What each plan costs a realtor, and when two seats make sense
List prices as of 1 November 2026; both vendors change plans often, so check the pricing pages before you buy.
| Level | ChatGPT | Claude |
|---|---|---|
| Free | $0; 3 file uploads a day | $0; up to 5 Projects |
| Individual paid | Plus, $20 a month | Pro, $20 a month or $17 a month billed annually |
| Heavy individual use | Pro from $100 a month | Max, $100 a month (5x Pro usage) |
| Team plan | Business: Standard seat $25 monthly or $20 billed annually, teams of 2 to 200 | Team: Standard seat $25 monthly or $20 billed annually, teams of 2 to 150 |
A solo agent should stay on an individual plan. Both team plans need at least two seats, so a one-person business would pay $40 to $50 a month for the privilege. That changes when you employ a transaction coordinator or an assistant. A team plan gives two things an individual plan does not: no training on your content by default, and, for Claude, shared Projects. Claude shares Projects only on Team and Enterprise; OpenAI says ChatGPT project sharing works on every plan including Free. If your assistant needs your listing style guide and your standard seller-update format, that sharing difference matters more than a few dollars. For a plan-by-plan look at the ChatGPT side alone, read the ChatGPT plan guide for realtors.
Usage caps are the part nobody can tabulate for you. Both vendors meter paid plans by usage windows, and the caps move. The cheap way to find out is to run your five tasks back to back on a paid plan and see whether you hit a wall. Document digests and market exports use more of your allowance than emails do, and Claude says file creation draws more from the limit than a normal chat.
Big files: what a 90-page inspection pack does in each tool
These figures come from each vendor's own help pages, fetched on 1 November 2026.
| Question | Claude | ChatGPT |
|---|---|---|
| Largest single file | 500 MB per file | 512 MB per file for documents |
| Length cap | PDFs up to 1,000 pages | Text and document files up to 2 million tokens each |
| Reads pictures inside a PDF | Text and visuals for PDFs of 100 pages or fewer; text only for 101 to 1,000 pages | Reads text and images in PDFs; the help page gives no page limit for the visual reading |
| Files at once | Up to 20 per chat | Up to 80 every 3 hours on paid plans; 3 a day on Free |
| Spreadsheets | CSV accepted; XLSX needs code execution and file creation switched on | CSV and spreadsheets up to about 50 MB |
| Single images | Up to 8000 by 8000 pixels | Up to 20 MB each |
| Files kept in a Project | 30 MB each, text extraction only (except multimodal PDFs) | 5 on Free, 25 on Go and Plus, 40 on Pro, Business and Enterprise |
Read that table for what it implies about a typical pack. A 90-page inspection report full of defect photos sits under Claude's 100-page visual limit, so Claude can look at the pictures. A 130-page disclosure bundle does not, so Claude would read only the text and ignore the photos. ChatGPT's help page says it can analyse images in PDFs but gives no page limit, which means you test it rather than assume. One more ChatGPT detail from the same page: files saved in a Project are handled differently from files attached to a chat, and embedded images may not be included, so attach the PDF directly to the chat when the pictures matter.
Then run a needle test. Pick one specific item deep in the pack, say a note on page 61 about staining under the roof flashing, and ask each tool: "What does the report say about the roof, and what does the photo on that page show?" Use the page numbers from your PDF viewer, not the numbers printed on the page. Claude's help centre gives that exact advice, and it applies to any tool, since scanned or combined PDFs rarely have printed numbers that match the file. A tool that answers about the roof but cites page 58 has read the document loosely, and you should not rely on its page references for anything a buyer will see.
Whatever either tool says about a defect is a reading aid for you, not a statement for the buyer. Repair costs from an assistant are guesses; a contractor's quote is the number. If a summary says the electrical panel is a concern, pass that to a licensed inspector and let them word it.
Market reports: give it the export and check the sums
The mistake to avoid is asking either assistant what the market is doing. Both can search the web, but the pages they find are rarely about your three streets. Give it the closed sales from your own system and ask for specific statistics. Before you upload an export from a multiple listing service, read your agreement with it: some licences restrict feeding listing data to third-party tools, and your brokerage may have its own rule.
Seven invented closed sales of three-bedroom homes make a quick check you can do by hand. The prices are $412,000, $425,000, $431,000, $438,000, $447,000, $460,000 and $475,000. The median is the fourth value, $438,000. The mean is the total, $3,088,000, divided by seven, which is $441,143 to the nearest dollar. Put a file like this through each tool and ask for both figures plus a line showing how it got them. If either answer differs from yours, you have learned something about that tool's arithmetic for free, before a wrong median goes into a seller update.
Attached is a CSV of closed sales: sale_date, address, beds, list_price, sale_price, days_on_market.
1. Report the median sale price, median days on market and the average sale-to-list ratio for 3-bed homes only.
2. Show the calculation steps, or the code you ran, so I can repeat them in a spreadsheet.
3. Name any row you excluded and why.
4. Write a three-sentence read for a seller. Do not describe anything the numbers do not show.Line two is the safety net. A tool that runs code on the file and shows it is easier to audit than one that reports a figure with no workings. ChatGPT limits data analysis on its free tier, and any free plan hits its upload caps quickly, so a paid plan matters more for this task than for emails.
Privacy defaults for an inbox full of other people's money
Consumer plans on both sides have a switch controlling whether your chats can be used to improve the models. OpenAI's switch sits under Data controls; Anthropic calls its version Model Improvement, and says that if you allow it, chats may be kept in de-identified form for up to five years. Anthropic's Incognito chats and ChatGPT's Temporary Chat are not used for training. On business plans, both vendors say they do not train on your content by default: ChatGPT Business and Claude Team. Check the labels in your own account, because the wording moves around.
The settings only answer the training question. They do not answer whether you should paste something in the first place. A working split for agents:
| Fine to paste | Keep out |
|---|---|
| Listing facts and the property address | A buyer's pre-approval letter or bank statements |
| Your open-house notes using first names only | Buyer or seller ID, date of birth, account numbers |
| Anonymised feedback ("buyer two disliked the kitchen") | A seller's mortgage balance or personal circumstances |
| Comparable sales from your own records | A listing-service export your licence says not to share |
Agent-style features raise the stakes. Anthropic's own safety page for its browser agent advises against using it on financial accounts or on sites holding other people's personal details, which describes a CRM exactly. Keep any assistant that can click around your accounts away from the system holding your clients, until you have read what it can do and who approves each action.
Teaching either assistant to sound like you
Both tools let you save standing instructions in a Project, which is where your voice belongs, so you stop retyping it. A filled-in version for an agent might look like this.
Project: Listings and client updates
Voice: plain and friendly, short sentences, first name terms, no exclamation marks.
Never describe the people who might live in a home. Never state that a price is wrong.
Seller updates: lead with the numbers I give you, then offer two options and ask which the seller prefers.
Emails: one next step per email, under 120 words, sign off with my first name only.
If a fact is missing, ask me instead of guessing.Here is an illustrative failure the instructions guard against, written for this tutorial rather than taken from either tool. An agent pastes three comparable listings and asks for a seller update. The draft ends: "Based on these comps, your home is overpriced by about $25,000 and should be reduced this week." That is a pricing verdict stated as fact, built from three listings, none of which share the seller's lot size. The better ending, the one the Project instructions above are written to steer towards, reads: "Three comparable homes are listed between $439,000 and $452,000; yours is at $465,000. Two options: hold for two more weeks, or reduce to $449,000. Which would you prefer?" The second version keeps the decision with the seller and uses only numbers you supplied.
Which assistant fits which part of the day
Use this as a list of what to test first, not as a verdict. I have not run these five tasks on both products for you, and anything claiming a universal winner is guessing.
| Part of the day | What decides it | Test first |
|---|---|---|
| Morning follow-ups | Voice match and invented details | Three emails from the same notes in each tool |
| Listing prep with photos | Whether it sticks to your facts | The prompt above, with the same 12 photos |
| Disclosure and inspection packs | Page limit, picture reading, page references | The page-61 needle test on a real pack |
| Market summaries | Arithmetic and showing workings | The seven-sale check, then your own export |
| Sharing with an assistant | Project sharing by plan | Claude needs Team; ChatGPT shares on any plan |
A decision procedure that fits on an index card: run the five tasks; log editing minutes per task; if the totals are within about 10 per cent, decide on file rules, sharing and privacy instead; if they are not, pick the lower total. Then keep paying for that one and drop the other to its free tier, as a spot-checker for anything important. Repeat the test when either vendor announces a change to plans or models, which happens more often than most tutorials admit.
The assistant is only one part of a realtor's stack. The CRM holds the contact history, and a receptionist or voice agent catches calls. For the CRM side, this CRM comparison for agents is the place to start. For deciding which jobs go to a human assistant and which to software, see the virtual assistant versus AI split. For a wider look at the same two products on listing and follow-up wording, there is also the listing, follow-up and limits comparison. If you want a wider look beyond real estate, Claude against ChatGPT for business writing covers proposals and reports.
Questions realtors ask before paying
Is a free plan enough for client emails and listing copy?
For drafting emails and listing descriptions from notes you type in, yes, as long as you keep buyer and seller identities out of the chat and the training setting is off. Free plans cap uploads and usage, so a heavy week of disclosure packs or market exports will push you to a paid plan within days.
Should I pay for both Claude and ChatGPT?
Only if your test shows each one doing better on a different task you do every week. Two $20 plans cost $40 a month, which is small next to an hour of your time, but most agents find one assistant plus the other's free plan as a spot-check is enough. Re-run the five-task test every few months because plans and limits change.
Can I upload a buyer's pre-approval letter?
Treat it as off limits on any personal plan. A pre-approval letter carries income and lending details the buyer shared with you for one purpose. If your brokerage allows AI tools on a business plan with no training by default, redact names and figures first and ask your broker before you upload anything.
Will either assistant know current prices in my area?
Not reliably. Both can search the web, but a handful of pages is a poor basis for a neighbourhood price claim. Give the assistant your own closed-sales export, ask for medians, and check the arithmetic in a spreadsheet before the number goes into a seller update or a social post.
Further reads
- AI Receptionist for Realtors: Cost, Call Routing and Showing Cover — Covers the phone side once your written follow-ups are sorted.
- Real Estate Chatbot Pricing: What Agents Pay for AI Lead Capture — Costs for website lead capture, which neither assistant does alone.
- Follow Up Boss Pricing: What a Small Team Really Pays — What a small team pays for the CRM that holds the follow-ups.
- AI Lead Generation for Real Estate Agents: No Bought Lists Needed — Lead sources to feed the emails you are now drafting faster.
- Best AI Tools for Commercial Real Estate Brokers, Compared by Job — If you also broker commercial space, tools by job.
- ChatGPT vs Claude vs Gemini vs Copilot for Small Business (2026) — A wider four-way comparison outside real estate.
- Rechat AI for Brokerages: What Lucy Does and What It Costs — Lucy's features from Rechat's own help pages, the pricing Rechat does not publish, and a 30-day pilot plan for a ten-agent office.
- AI Marketing for Real Estate Agents: A Four-Week Content Plan — Week by week: a listing launch, an area guide, a market update from your own figures and a proof week, each turned into posts, a script and an email.
- Best Free AI for Realtors: Three Jobs, the Limits and When to Pay — Three jobs a realtor can do on free AI plans, a fair-housing check for every output, and the signs that a paid plan is worth it.
- ChatGPT Training for Real Estate Agents: A One-Week Team Plan — A team leader's one-week plan for training agents on ChatGPT: daily sessions, a rules sheet, five workflows and an adoption check.
- ChatGPT for Realtors: Brokerage Rules for Emails, Disclosures, Follow-Ups — A brokerage owner's rulebook for agent ChatGPT use: what may be summarised, what must be redacted, the fact-check pass and five reusable prompts.
- Real Estate Agent Recruiting With AI: Outreach and Onboarding Plan — A four-touch recruiting sequence, an onboarding checklist and a first-30-days plan for a brokerage, with the compliance lines AI must not cross.
- Can Realtors Use HeyGen Avatars for Listings? Rules and Setup — How a realtor can put their own HeyGen avatar on listing updates, what to disclose, and how to avoid MLS, brokerage and fraud trouble.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: claude.com/pricing, openai.com/business/pricing, OpenAI's file-uploads FAQ, Claude's file-upload help article, Anthropic and OpenAI privacy and data-control pages (checked 1 November 2026).