Neither is clearly better for an agent, so test both on your own listing. Give each the same facts, the same three tasks and the same banned-wording list, then pick the tool whose drafts need less editing. Claude and ChatGPT both handle listing copy and follow-up emails well; the differences are voice, invention and limits.
The test below uses three tasks an agent actually does every week: a listing description, a follow-up to a buyer after a viewing, and a progress update to a vendor. Each has a prompt you can paste, an illustrative output, and a short list of what to fix. Whichever tool wins, a fair-housing wording check runs over every listing before it goes live, because the published words are yours, not the tool's.
Set up the test so the comparison is fair
Pick one real listing you know well and strip anything private: the vendor's name, the exact street number and any detail they asked you to keep quiet. Write ten facts about it in plain lines. Facts matter more than prompts: both tools will fill silences with plausible detail, so the less you leave unsaid the less they invent.
Property facts (illustrative):
- Three-bedroom semi-detached house, built 1960s
- 1,150 sq ft, south-facing garden about 40 ft long
- New boiler fitted last year, double glazing throughout
- Kitchen refitted five years ago, no dishwasher
- Off-street parking for two cars
- Ten-minute walk to the primary school and the station
- Loft partly boarded, no conversion
- Bathroom is original and dated
- Asking price $425,000
- Vendors want a quick sale, chain-freeUse identical facts for both tools. If you pay for neither, use the free plans; the task is small. Keep each tool's reply and score it afterwards. Allow an hour for all three tasks in both tools.
Task one: the listing description
Write a listing description of 150 to 180 words for the property
below. Use only the facts given; do not add any feature, distance or
condition I have not stated. Mention the dated bathroom honestly.
Tone: warm, plain, no clichés (no "stunning", "must-see",
"nestled"). Do not describe who the home would suit; describe the
home. End with the price. Facts: [paste the ten facts]Good output opens with the strongest true fact, gives the layout, then the practical points, and mentions the bathroom without apologising. A typical weak output either hides the dated bathroom or invents "a sun-drenched kitchen" you never mentioned. Score each version on four things: facts correct (count the slips), invented features (count them), banned words (count them) and edits you would make (count the minutes).
What you usually fix
Three recurring fixes: a sentence that exaggerates ("an exceptional kitchen" for a five-year-old refit), a distance described as a "short stroll" when you said ten minutes, and a closing line addressed to a type of buyer. Strip all three, then compare how many each tool produced. A tool that gives you 160 words needing two edits has saved real time; one that gives 200 words needing eight has not.
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 nowTask two: the buyer follow-up after a viewing
This tests tone and structure. Use a viewing note, not a name.
Write a follow-up email of under 120 words to a buyer, [first name],
after a second viewing of the three-bedroom house. They liked the
garden and the parking, worried about the dated bathroom, and said
they need to sell their own flat first. Offer two things: a
bathroom quote range I can arrange, and a call on Thursday. No
pressure, no invented urgency, no mention of other buyers. Sign off
as [your name].Illustrative good output: a greeting, one line on what they liked, one line acknowledging the bathroom and offering to arrange a quote, one line on their sale timing, a clear call-to-action for Thursday. What to look for in the weaker version: a made-up reference to "interest from other parties", a slightly pushy "don't miss out", or a claim about the bathroom cost you never gave. Delete them and note the tool.
Because follow-up emails go out many times a week, the useful measure is the first-draft keeper rate: out of five drafts on five different viewing notes, how many would you send after one small edit? Run five, not one. A tool that gets 4 out of 5 is worth building into your routine; how to automate buyer and vendor follow-up with AI covers how to make that routine repeatable.
Task three: a vendor update that is honest about slow weeks
Vendor updates are harder than they look, because the news is often bad and the tool's instinct is to sound upbeat. Give it numbers.
Write a vendor update email of under 150 words. Facts: listed
three weeks ago; 14 enquiries, 6 viewings, 0 offers; two viewers
said the bathroom put them off; a similar house on the same road
reduced by 3% after four weeks. Suggest two options: a modest price
review, or a bathroom credit offered to buyers. Be factual and
respectful. Do not promise an outcome. Sign off as [your name].Check the arithmetic and the claims. A 3 per cent reduction on $425,000 is $12,750, so a price review to $412,250 is a figure you might offer as an example; the tool should not invent it unless you ask. Score for honesty: does the draft say "no offers yet" plainly, does it avoid blaming the vendor, and does it keep promises out? A draft that spins zero offers as "strong early interest" fails regardless of how polite it is.
The fair-housing wording check
Whatever the tool, the published text is yours. Rules differ by place, so check your own brokerage and local guidance; the practical principle is the same everywhere: describe the property, not the people who might live in it. Build a banned list into every listing prompt and scan every output against it.
| Risky wording (illustrative) | Why it is risky | Property-focused alternative |
|---|---|---|
| "Perfect for a young family" | Describes the buyer, not the home | "Three bedrooms, garden about 40 ft long" |
| "Ideal for professionals" | Signals a preferred type of occupant | "Ten minutes' walk to the station" |
| "Quiet, mature neighbours" | Characterises the people nearby | "Residential cul-de-sac with off-street parking" |
| "Close to a church" or "safe area" | Can imply who lives or belongs there | State distances to named amenities and leave out judgements |
| "Walk-up, not suitable for the elderly" | Excludes by implication | "Second-floor flat, 24 steps, no lift" |
Two practices make this safer. First, ask each tool to rewrite the description and list every phrase that refers to a person or group, then check that list yourself; a tool that returns "none" is giving you a signal, but it can miss things, so read the text too. Second, keep a before and after in your file when a phrase is removed, so you can show what you did if a question arises. A broader checklist for tools that talk to leads is in the leasing chatbot fair-housing checklist.
Three more tests once the first three pass
Rewrite a tired listing without changing the facts
Many agents start with an existing description that reads like every other one. Paste it in and ask each tool to rewrite it at the same length, keep every fact, remove clichés and flag anything it changed. Illustrative before: "A stunning, must-see family home nestled in a sought-after location, offering spacious accommodation throughout." After, from a good draft: "Three bedrooms over 1,150 sq ft, a south-facing garden about 40 ft long and parking for two cars. The kitchen was refitted five years ago; the bathroom is original." Check the "flag anything changed" list. A tool that reports "I replaced 'spacious' with the square footage" is behaving well. One that silently drops the parking line has cost you a selling point.
Short versions for portals and social posts
Ask for a 40-word portal summary, a 25-word caption for a post and three subject lines for an email alert, all from the same facts. This is a cheap test of whether the tool holds the facts when squeezed. Typical slips are rounding "1,150 sq ft" to "over 1,200", or converting a ten-minute walk into "moments from". Compare the short versions with your fact list line by line. Shorter text hides errors less well than long text, which is why this check takes under a minute per version.
The tricky reply: a buyer who has gone quiet
Give both tools this: "Buyer saw the house twice, then stopped replying for nine days. Write a message of under 60 words that asks if they would like to hear about the quote for the bathroom or would rather I stop sending updates. No guilt, no urgency." The better draft gives the buyer an easy way to say no. The worse draft invents a reason to reply ("another offer may be coming"). Treat any invented pressure as a failure, because it is dishonest and usually breaks local advertising or conduct rules too.
What five drafts per task looks like in a table
One draft proves nothing, so run five per task per tool and fill a table like the illustrative one below. The numbers show a method, not a result.
| Task | Tool | Facts wrong (of 5 drafts) | Invented content | Banned phrases | Average edit minutes | Sendable after one edit |
|---|---|---|---|---|---|---|
| Listing description | Tool A | 1 | 2 | 1 | 4 | 4 of 5 |
| Listing description | Tool B | 0 | 3 | 0 | 5 | 3 of 5 |
| Buyer follow-up | Tool A | 0 | 1 | 0 | 2 | 5 of 5 |
| Buyer follow-up | Tool B | 0 | 0 | 1 | 3 | 4 of 5 |
| Vendor update | Tool A | 1 | 1 | 0 | 5 | 3 of 5 |
| Vendor update | Tool B | 0 | 1 | 0 | 4 | 4 of 5 |
Read the table by column, not by row. Invented content is the one to weight most. In this made-up example Tool B has fewer wrong facts but invents more on listings, which is a trade-off you would only see by counting. Tool A saves more time on follow-ups, the task you do most. A sensible reading might be to use Tool A for follow-ups and keep an eye on its vendor updates, but accept the added cost of two tools only if the saving justifies it. Most agents settle on one tool and learn its habits.
A different property, to see if the result holds
One listing is a sample of one. Before you commit, rerun the listing task on a property that looks nothing like the first, such as a one-bedroom flat with a lift, a service charge of $210 a month, 84 years left on the lease and no parking. Flats test different things: the tool must report the service charge and lease length exactly, not round them, and must not say "low running costs" because that is a judgement. Illustrative good opening: "One-bedroom flat on the third floor with a lift. Service charge $210 a month; 84 years remain on the lease." A weaker one tucks the lease length into a vague "long lease". Buyers and lenders care about that figure, so a tool that softens it is a risk in your marketing, not a style quirk.
Do the same with the follow-up: write a note for a buyer who is paying cash and one who needs a mortgage, and see whether the tool keeps its hands off financial advice. A reply that says "you should easily get approved" has crossed a line; one that offers to share a list of the documents lenders usually ask for has not. These small differences, rather than the headline comparison, tend to settle which tool you trust with your name on the message.
Where each task goes wrong, and how it shows up
- Invented amenities. The listing mentions "a utility room" or "fitted wardrobes" that you never listed. It shows up when a buyer asks about them at the viewing. Fix: tell the tool to ask rather than add, and check every noun against your fact list.
- Soft numbers. "Around 1,200 sq ft" from 1,150, "close to the station" from ten minutes. It shows up in portal listings where figures are checked. Fix: instruct "copy numbers exactly".
- Manufactured urgency. "Don't miss out" or "interest is high". It shows up as a vendor asking why you said there was interest. Fix: ban urgency language in the project instructions.
- Tone drift in long threads. After many edits, the voice changes. It shows up as a draft that no longer sounds like you. Fix: start a new chat inside the project for each new property.
- Pasted private details. A vendor's phone number or reason for selling ends up in a chat. Fix: placeholders from the start, and delete the chat when the job ends.
If you want a second pair of tools for the follow-up itself, the comparison of AI CRMs for real estate agents shows which ones send the messages for you; the assistants tested here only draft.
Plans, privacy and limits that affect the choice
Figures are list prices checked on 4 November 2026; confirm them on the vendor pages before buying.
| Point | Claude | ChatGPT |
|---|---|---|
| Start free | Free plan, usage limits, up to five projects | Free plan: unlimited text chats, but limited uploads and tools; projects of up to 5 files |
| Solo paid plan | Pro: $17 a month annually, $20 monthly | Plus: $20 a month (Go, at $8, sits below it) |
| Team plan for an assistant | Team Standard: $20 a seat annually, $25 monthly, minimum two seats | Business: $20 a seat annually, $25 monthly, minimum two users |
| Holding your voice and templates | Projects with instructions and files (30 MB per file); sharing needs Team or Enterprise | Projects with instructions and files (25 files per project on Go and Plus); sharing on every plan |
| Training on your content | Team and Enterprise: none by default; consumer plans: a setting you manage | Business and Enterprise: none by default; consumer plans: opt-out through "Improve the model for everyone" |
The limits that matter in practice are not price. They are these. Neither tool knows your market: any comparable sale, price trend or school catchment it mentions is unchecked unless you supplied it. Neither sees your listings system, so you paste facts in. Both can produce confident wording on legal points such as disclosures or contract terms, which belong to your broker or lawyer. And both can vary between runs, so a draft you liked yesterday may not reappear today; save the good ones.
Keep your voice in a project instead of pasting it every time
After the test, make a project in the tool you picked. Put in three of your own past descriptions you are proud of, your banned-word list, your sign-off, and a line such as "never mention a person or group the home suits". Now every new chat starts with your style and your rules. If you share the project with an assistant (in Claude that needs a Team plan), they get the same behaviour. Claude Projects for marketing: brand voice, briefs and a content calendar shows the pattern in more detail, and the same idea works in ChatGPT Projects.
A filled-in project instruction
You write for a small real estate agency. Voice: warm, plain,
specific. Use only facts I give. Never add features, distances or
conditions. Never describe who a home suits. Banned words: stunning,
must-see, nestled, gem, perfect for. Listings: 150 to 180 words, end
with the price. Emails: under 120 words, one clear next step, no
invented urgency. If a fact is missing, ask me instead of guessing.The last sentence is the useful one. A tool told to ask before guessing asks about the boiler's age rather than making one up, and that single behaviour removes most invented detail.
A single sum to decide whether it is worth paying
Say an agent writes six listing descriptions, 25 follow-up emails and eight vendor updates in a typical month. Suppose drafting each by hand takes 15, 6 and 12 minutes. That is 90 plus 150 plus 96, or 336 minutes, about 5.6 hours. If the better tool leaves you with drafts needing a third of that time to edit, the saving is roughly 3.7 hours a month. On a $20 plan, that is under $6 an hour of time saved; on a $25 team seat it is a little more. These figures are illustrative; run the test to find your own. The point is that the subscription is small beside the cost of one dull listing that sits for a week longer than it should.
Score the three tasks and decide
Use a simple sheet: three tasks, two tools, columns for facts wrong, invented content, banned phrases and edit minutes. Add a final column: would I send this? Total the "would I send" count over five drafts of each task, not one. If both tools score within one point, choose on convenience: which is already on your phone, which your assistant prefers, which has the sharing features you need. If one scores clearly higher on invented content, choose that one, because invention is the error that causes trouble later.
Revisit the choice in three months. Models change, plans change and your own needs change as you add more routines. If you would like help picking the first routine to automate and setting up the project, that is what my AI implementation consultation covers, but the test above gives you the answer without it.
Questions agents ask about AI for listings and follow-up
Which is better for listing descriptions?
Neither wins every time. Run your own listing through both with the same prompt and compare facts, rhythm and anything invented. One tool will usually sit closer to your voice once you give it two or three of your own past descriptions as examples.
Can AI wording break fair-housing rules?
Yes. A tool can drift into phrases that describe the kind of person a home suits rather than the home itself. Ban those phrases in the prompt, then read every description yourself before publishing. Your brokerage or association guidance has the final word.
Should I paste client names and offers into the tools?
Use first names or placeholders and leave out phone numbers, financial details and anything the client told you in confidence. Business plans do not train on your content by default, but placeholders cost nothing.
Do I need a paid plan?
Not to start. Free plans handle all three tasks. Pay when you hit upload or usage limits or want more room for project files. In Claude, sharing a project with an assistant needs a Team or Enterprise plan; ChatGPT allows project sharing on every plan.
Further reads
- Follow Up Boss Pricing: What a Small Team Really Pays — What a follow-up CRM really costs a small team
- Real Estate Chatbot Pricing: What Agents Pay for AI Lead Capture — What agents pay for AI lead capture
- Virtual Assistant or AI for a Real Estate Agent: What Goes Where — What goes to a person and what to AI
- Real Estate AI Chatbot Demo: A Test Checklist Before You Sign — Questions to ask before buying a chatbot
- AI Voice Agent for Real Estate Follow-Up: Setup, Scripts, TCPA — Rules for calling and texting leads
- Real Estate AI Chatbot Features: A Checklist Before You Buy — Features to check before buying a chatbot
- Claude or ChatGPT for Realtors: Which Assistant Fits Your Day? — Five realtor tasks to run on both assistants, what each plan costs, how each reads a 90-page disclosure pack, and the privacy settings that matter.
- AI Sales Role-Play for Solo Real Estate Agents: Free Tools Compared — Which free AI tools can play a seller or buyer, what each free tier allows, and an objection bank and weekly routine to practise with.
- Open House Sign-In Sheet With AI: Follow-Up Within the Hour — A printable sign-in sheet, consent wording, buyer-type message templates and a 60-minute plan for personal follow-up after every open house.
- Can ChatGPT Write Real Estate Listings That Pass Fair Housing? — A listing-copy workflow with an edit-pass prompt, before-and-after wording and a human fair-housing check.
- 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, chatgpt.com/pricing and OpenAI's business pricing page (checked 4 November 2026); OpenAI's help article on projects; Anthropic's help articles on projects and file uploads.