{"id":40,"date":"2026-09-27T12:40:05","date_gmt":"2026-09-27T12:40:05","guid":{"rendered":"https:\/\/borstev.com\/blog\/?p=40"},"modified":"2026-09-27T12:49:15","modified_gmt":"2026-09-27T12:49:15","slug":"gpt-6-astra-real-estate","status":"publish","type":"post","link":"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\/","title":{"rendered":"GPT-6 Astra for Real Estate: I Gave It a Messy CRM and Watched It Break"},"content":{"rendered":"\n<div class=\"wp-block-group is-style-plain\"><div class=\"wp-block-group__inner-container is-layout-flow wp-block-group-is-layout-flow\">\n<figure class=\"wp-block-embed is-type-video is-provider-youtube wp-block-embed-youtube wp-embed-aspect-16-9 wp-has-aspect-ratio\"><div class=\"wp-block-embed__wrapper\">\n<iframe loading=\"lazy\" title=\"I Tested GPT-6 Astra for Real Estate: What Broke in 20 Minutes\" width=\"500\" height=\"281\" src=\"https:\/\/www.youtube.com\/embed\/KojLwJdgd8o?feature=oembed\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" referrerpolicy=\"strict-origin-when-cross-origin\" allowfullscreen><\/iframe>\n<\/div><\/figure>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"wp-block-paragraph\">I gave GPT-6 Astra twenty minutes inside a copy of a real estate CRM, with fake leads, dead records and no clean notes, and asked it to follow up while I was busy. Within the first run it confirmed a showing nobody approved, told a lead a property was available when that unit had gone under contract the same morning, messaged a dead phone number, and logged a booking that never happened.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The model is not broken. I ran the same prompt twice more, once after adding guardrails and once after cleaning the CRM, and by the third run it did the job correctly. What follows is what changed between those runs, and the one setting that did about half the work.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">If you watched my video on the five questions to ask before deploying an AI agent, this is the sequel. Same five questions, new model. I am not going to re-teach the framework here, you can <a href=\"https:\/\/borstev.com\/blog\/ai-agent-for-real-estate-5-questions\/\">check the five questions in full<\/a>. This piece is about what actually broke when I ran a GPT-6 Astra real estate setup for real.<\/p>\n\n\n\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_88 counter-hierarchy ez-toc-counter ez-toc-grey ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title\" style=\"cursor:inherit\">Table of Contents<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"Toggle Table of Content\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Toggle<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #999;color:#999\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewBox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #999;color:#999\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewBox=\"0 0 24 24\" version=\"1.2\" baseProfile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1 ' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\/#What_GPT-6_Astra_is_and_where_to_find_it\" >What GPT-6 Astra is, and where to find it<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\/#Test_one_what_happens_with_no_boundaries\" >Test one: what happens with no boundaries<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\/#The_one_setting_that_did_half_the_work\" >The one setting that did half the work<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\/#Test_two_what_happens_when_the_data_is_a_mess\" >Test two: what happens when the data is a mess<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\/#What_the_three_runs_looked_like_side_by_side\" >What the three runs looked like side by side<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\/#The_legal_part_including_a_2026_change_most_coverage_missed\" >The legal part, including a 2026 change most coverage missed<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-7\" href=\"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\/#Test_three_drift_and_why_launch_day_is_not_the_finish_line\" >Test three: drift, and why launch day is not the finish line<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-8\" href=\"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\/#Where_your_judgment_has_to_stay\" >Where your judgment has to stay<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-9\" href=\"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\/#The_setup_checklist\" >The setup checklist<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-10\" href=\"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\/#FAQ\" >FAQ<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-11\" href=\"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\/#Get_the_checklist\" >Get the checklist<\/a><\/li><\/ul><\/nav><\/div>\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"What_GPT-6_Astra_is_and_where_to_find_it\"><\/span>What GPT-6 Astra is, and where to find it<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">OpenAI released GPT-6 Astra on 3 September 2026, first to a limited set of organisations and then over the following days to Plus, Pro, Business and Enterprise users, along with the API and AWS.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">One practical note if you are on Plus and went looking for it in your normal chat window without success. Astra lives inside ChatGPT Work and Codex, not the standard chat interface. On Enterprise it is off by default at launch, so an admin has to switch it on before anyone in your office can touch it.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The more important thing is how OpenAI positions it. They are not selling better writing. <a href=\"https:\/\/openai.com\/index\/gpt-6-astra\/\" target=\"_blank\" rel=\"noopener\">Their own announcement<\/a> frames Astra around computer use: it can operate the same applications you use daily, including ones with no integration available, by reading what is on screen and clicking through it. It sends. It books. It updates records.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Which changes the question you should be asking. For three years the question with ChatGPT was what to prompt. With an agent that can click buttons in your CRM, the question becomes what you are allowing it to touch. Permissions now matter more than prompt wording.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Test_one_what_happens_with_no_boundaries\"><\/span>Test one: what happens with no boundaries<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">I copied a CRM, filled it with fake leads, and gave Astra exactly three tools: read contacts, send a message, book a calendar slot. Nothing else.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">First run, my instruction was two words long. Help these leads.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">It helped. It also went considerably further than I wanted. It confirmed a showing time on its own authority. It told one lead a property was still available, and in my test data that unit had gone under contract that morning. Inventory in this business changes daily and sometimes hourly, so a confident answer about availability from something that cannot see your latest status is a trust problem waiting to happen, and in some situations a legal one.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Then came the failure that made the point for me. I asked it to follow up with everyone who had asked about Elm Street. Elm Street had two contacts with the same name and two different numbers, one current and one long dead. Astra picked the dead one, sent the message, and then logged &#8220;booked showing&#8221; in the record. Nothing had been booked. It filled the gap with a guess.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">You already know what happens to leads nobody answers quickly. They call the next agent. What surprised me is that a badly configured agent produces the same outcome, just faster and with your name on it.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"The_one_setting_that_did_half_the_work\"><\/span>The one setting that did half the work<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">I reset everything and ran it again, this time writing the boundaries before the goal. Gather interest, answer general questions, and never commit to price, availability or scheduling without a human confirming first.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Same model, same CRM, different outcome. Astra did not get smarter between those two runs. It got clearer borders.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Then I turned on the setting that matters most, and this one is worth doing before you connect anything. In ChatGPT Work, enable <strong>confirmation policies<\/strong>. OpenAI describes these in <a href=\"https:\/\/openai.com\/index\/gpt-6-astra-next-generation-work\/\" target=\"_blank\" rel=\"noopener\">their enterprise documentation<\/a> as safeguards that &#8220;can require approval before consequential actions,&#8221; alongside automated review of unsafe or unauthorised tool calls. In practice it means Astra stops and asks you before it sends a message or books a slot.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">On Business and Enterprise plans your admin can go further and restrict Astra to approved websites and desktop applications, control uploads and downloads, and manage browsing history.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Turn these on before you connect your CRM. Not after something breaks.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">I have made the opposite mistake myself, on one of my own agents. I told it precisely what to search for and never told it what to skip. It did exactly what I asked and I ended up paying for a pile of data I could not use. The AI performed correctly. I wrote a bad brief.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">So rule one for any GPT-6 Astra real estate deployment: write down what it must never do, before you write down what it should do. If you cannot list the never-dos, you are not ready to connect it.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Test_two_what_happens_when_the_data_is_a_mess\"><\/span>Test two: what happens when the data is a mess<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">An agent works with what it can see. Give it a messy database and it has to guess, and every guess is a chance to embarrass you.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The messy version of my test CRM had the four problems I see in almost every agency I audit. Duplicate records, because the same person came in from Zillow, then a Meta form, then a referral. No consent trail. No closed-lost status, so the file still shows an active buyer who actually purchased eight months ago. No real notes, because the context is sitting in a call nobody logged.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Astra contacted the same person twice in two days using two different tones, formal on Monday and casual on Tuesday. To that lead it reads like nobody in the office is paying attention.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">There is a subtler failure underneath that one. When you are showing the next house, you do not remember that this particular lead said &#8220;not now, call me in six months.&#8221; Astra does not remember either, unless somebody put it in a field. All it sees is no reply, so it pushes. That is how you burn a good lead by being annoying rather than by being slow.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Then I cleaned the CRM. Merged the duplicates, marked who had already bought, marked do-not-contact, and added one line of last context per record. Same model, same prompt, no changes to the instructions at all.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This time it skipped the buyer. It skipped the leads with no consent logged. It sent four correct follow-ups with the right name attached to the right property.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Same brain. Better fuel. If your own database is the bottleneck here, I wrote a full walkthrough of the cleanup in this piece on reviving dead CRM leads, and that audit is the prerequisite for everything on this page.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"What_the_three_runs_looked_like_side_by_side\"><\/span>What the three runs looked like side by side<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1017\" height=\"610\" src=\"https:\/\/borstev.com\/blog\/wp-content\/uploads\/2026\/09\/gpt-6-astra-real-estate-crm-test-results.png\" alt=\"\" class=\"wp-image-41\" srcset=\"https:\/\/borstev.com\/blog\/wp-content\/uploads\/2026\/09\/gpt-6-astra-real-estate-crm-test-results.png 1017w, https:\/\/borstev.com\/blog\/wp-content\/uploads\/2026\/09\/gpt-6-astra-real-estate-crm-test-results-300x180.png 300w, https:\/\/borstev.com\/blog\/wp-content\/uploads\/2026\/09\/gpt-6-astra-real-estate-crm-test-results-767x460.png 767w\" sizes=\"auto, (max-width: 1017px) 100vw, 1017px\" \/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"The_legal_part_including_a_2026_change_most_coverage_missed\"><\/span>The legal part, including a 2026 change most coverage missed<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Before going further, the obvious disclaimer, and I mean it: I am not a lawyer, this is not legal advice, and you should talk to your own attorney about your specific setup and the states you operate in.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">With that said, here is what I would want to know if I were running an agency.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">When AI texts or calls your leads, US telemarketing law is in play. The FCC&#8217;s February 2024 Declaratory Ruling confirmed that AI-generated voices count as &#8220;artificial or prerecorded voice&#8221; under the TCPA, with no exemption for conversational AI or large language model agents. Statutory damages run from $500 to $1,500 per call, with no aggregate cap, which is what makes a high-volume campaign genuinely dangerous rather than merely risky.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The part that most AI-for-real-estate content has not caught up with yet is that the consent standard moved in 2026. On 25 February 2026 the Fifth Circuit decided <em>Bradford v. Sovereign Pest Control of Texas, Inc.<\/em>, holding that the TCPA&#8217;s actual statutory text requires only prior express consent rather than the heightened prior express written consent the FCC has required since 2012. That ruling binds Texas, Louisiana and Mississippi. In the other 47 states the FCC&#8217;s written-consent framework still applies.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Two other details worth carrying into any setup. Since April 2025, consumers can revoke consent in any reasonable manner, including words like stop, quit, end, revoke, opt out, cancel or unsubscribe, and you are expected to honour it within ten business days. And an existing business relationship does not substitute for consent when an artificial voice is involved, which catches out a lot of people who assume their past-client list is fair game.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The practical rule I use: if the record cannot show when and how that person agreed to be contacted, the agent does not touch them until your attorney says otherwise.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Test_three_drift_and_why_launch_day_is_not_the_finish_line\"><\/span>Test three: drift, and why launch day is not the finish line<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Treat an agent like a new hire rather than a finished product. Check it weekly or every two weeks, not whenever something visibly breaks, because by the time it visibly breaks it has already cost you.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">To test this I trained the agent for a slow market. Homes sitting sixty days, relaxed cadence, plenty of &#8220;just checking in, no pressure, here is a market update.&#8221; In a slow market that tone is correct.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Then I switched the data feed to a hot week. Homes moving in seven days, multiple offers, listings gone within hours. Nobody retrained the agent.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">One exchange showed the whole problem. A lead asked whether 42 Maple was still available on Tuesday. Astra answered with the slow-market script: great question, I can send you similar homes this week. In a sixty-day market that is a perfectly good reply. In a hot one, that buyer had already booked with the agent who said yes, Tuesday at four is open, want me to hold it pending confirmation.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">So write down what working actually means, in numbers you can check. Replies within five minutes, with correct information, correctly distinguishing buy intent from sell intent at least 95% of the time. Then name a person and name a day. Every Monday, fifteen minutes, review ten conversations, check the bookings, check the opt-outs.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Where_your_judgment_has_to_stay\"><\/span>Where your judgment has to stay<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">There is one job I gave Astra that it did quickly, confidently, and in a way that could put an agency in serious trouble.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">I asked it to recommend listings to buyers based on what similar past clients had liked. On the surface that is ordinary personalisation. Underneath, &#8220;similar clients&#8221; in historical housing data is not a neutral signal. Decades of housing segregation are baked into who bought where. An agent optimising on that pattern can start steering buyers toward or away from neighbourhoods in ways that track race, without anybody typing a word about race and without the model deciding to discriminate. It found a correlation and optimised for it.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">HUD&#8217;s guidance on AI in housing is clear that the Fair Housing Act applies to automated systems exactly as it applies to a person. Same disclaimer as before, talk to your attorney about what applies where you operate, but this is precisely the behaviour fair housing law exists to catch.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The safe version is narrow. Recommend only from what the buyer told you: budget, bedrooms, commute, must-haves like a yard or a garage. Write it into the prompt in plain words: <em>recommend only from the buyer&#8217;s stated criteria, never use past clients, demographics, or neighbourhood profiles.<\/em> When a buyer asks about schools or areas, point them to neutral public sources and let them draw their own conclusions.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Then list the decisions that stay with a human. Price, steering, availability promises, anything touching a contract. The agent gathers interest, answers general questions, and brings in a person before any commitment.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"The_setup_checklist\"><\/span>The setup checklist<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Step<\/th><th>What to do before you connect Astra to anything<\/th><\/tr><\/thead><tbody><tr><td>Write the never-do list<\/td><td>Price, availability, scheduling, contract talk. If you cannot list these, stop here<\/td><\/tr><tr><td>Turn on confirmation policies<\/td><td>In ChatGPT Work, before the CRM connection, not after<\/td><\/tr><tr><td>Restrict the app list<\/td><td>On Business and Enterprise, have your admin limit Astra to approved apps<\/td><\/tr><tr><td>Clean the database<\/td><td>Merge duplicates, mark bought, mark do-not-contact, one line of context per record<\/td><\/tr><tr><td>Check the consent trail<\/td><td>No record of when and how they agreed means the agent does not contact them<\/td><\/tr><tr><td>Define working in numbers<\/td><td>Reply time plus an accuracy rate you can actually measure<\/td><\/tr><tr><td>Name a drift owner<\/td><td>A person and a weekday. Fifteen minutes, ten conversations, bookings, opt-outs<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"FAQ\"><\/span>FAQ<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Is GPT-6 Astra safe to use on a real estate CRM?<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Only with guardrails set first and clean data underneath. In my own test, Astra with no boundaries confirmed a showing on its own, claimed a sold property was available, and logged a booking that never happened. The same model with a written never-do list, confirmation policies enabled and a cleaned database sent four correct follow-ups and correctly skipped leads who had already bought or had no consent recorded.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Where do I find GPT-6 Astra if I am on ChatGPT Plus?<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Inside ChatGPT Work and Codex rather than the standard chat window. On Enterprise plans access is off by default at launch, so an administrator needs to enable it for your workspace first.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>What are confirmation policies in ChatGPT Work?<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A safeguard OpenAI describes as requiring approval before consequential actions, such as sending a message or booking a calendar slot, along with automated review of unsafe or unauthorised tool calls. Turn them on before connecting a CRM rather than after something goes wrong.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Do I need consent before an AI agent calls my real estate leads?<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Generally yes in the US. The FCC&#8217;s February 2024 ruling treats AI-generated voices as artificial voices under the TCPA, and damages run $500 to $1,500 per call with no cap. A February 2026 Fifth Circuit decision lowered the standard to prior express consent in Texas, Louisiana and Mississippi, while the FCC&#8217;s written-consent framework still governs the other 47 states. This is not legal advice, so confirm your own position with an attorney.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Can an AI agent create a Fair Housing problem without anyone intending it?<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Yes. An agent recommending listings based on what similar past clients chose can reproduce historical segregation patterns and effectively steer buyers by neighbourhood, with nobody mentioning race at any point. HUD&#8217;s guidance confirms the Fair Housing Act applies to automated systems the same as to people. Restrict recommendations to the buyer&#8217;s own stated criteria.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>How often should I check an AI agent after launch?<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Weekly or every two weeks, with a named owner. Agents trained during a slow market keep that pace after the market turns hot, and the resulting slow replies look unresponsive to buyers who have already booked elsewhere.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Get_the_checklist\"><\/span>Get the checklist<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">This is the audit we run with every agency before we install anything, along with the never-do list I use as a starting template. Send us an email to bogomil@borstev.com and will send it to you.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">We build, configure and maintain AI receptionists for real estate businesses. They answer the phone, handle questions, qualify leads and book them into your calendar, with guardrails built in: no price promises, no availability promises, and a human before any commitment. Before any installation we audit the database first.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">If you want us to run that audit on yours, <a href=\"https:\/\/calendly.com\/bogomil-stoev\/custom-ai-agents\" target=\"_blank\" rel=\"noopener\">book a call<\/a>.<\/p>\n<\/div><\/div>\n\n\n\n<script type=\"application\/ld+json\">\n{\n  \"@context\": \"https:\/\/schema.org\",\n  \"@graph\": [\n    {\n      \"@type\": \"Article\",\n      \"headline\": \"GPT-6 Astra for Real Estate: I Gave It a Messy CRM and Watched It Break\",\n      \"description\": \"A first-hand test of GPT-6 Astra inside a real estate CRM: what broke with no guardrails, what changed after enabling confirmation policies, and what changed again after cleaning the data.\",\n      \"author\": {\n        \"@type\": \"Person\",\n        \"name\": \"Bogomil Stoev\",\n        \"url\": \"https:\/\/borstev.com\/about\",\n        \"sameAs\": [\n          \"https:\/\/www.linkedin.com\/in\/bogomils\/\",\n          \"https:\/\/www.youtube.com\/@borstev\"\n        ],\n        \"jobTitle\": \"AI Agent Consultant\",\n        \"knowsAbout\": [\"AI agents\", \"GPT-6 Astra\", \"Real estate lead generation\", \"CRM automation\", \"AI agent deployment\"]\n      },\n      \"publisher\": {\n        \"@type\": \"Organization\",\n        \"name\": \"Borstev\",\n        \"url\": \"https:\/\/borstev.com\"\n      },\n      \"datePublished\": \"[2026-09-27]\",\n      \"dateModified\": \"[2026-09-27]\",\n      \"mainEntityOfPage\": \"https:\/\/borstev.com\/blog\/gpt-6-astra-real-estate\",\n      \"about\": {\n        \"@type\": \"SoftwareApplication\",\n        \"name\": \"GPT-6 Astra\",\n        \"applicationCategory\": \"BusinessApplication\",\n        \"creator\": {\"@type\": \"Organization\", \"name\": \"OpenAI\"}\n      }\n    },\n    {\n      \"@type\": \"VideoObject\",\n      \"name\": \"I Tested GPT-6 Astra for Real Estate: What Broke in 20 Minutes\",\n      \"description\": \"I tested gpt 6 astra for Real Estate inside a test CRM - read contacts, send follow-up, book calendar. It was fast, confident, and it broke 3 times in 20 min.\",\n      \"thumbnailUrl\": \"https:\/\/i.ytimg.com\/vi\/KojLwJdgd8o\/maxresdefault.jpg\",\n      \"uploadDate\": \"[YYYY-MM-DDT00:00:00+03:00]\",\n      \"duration\": \"PT[M]M[S]S\",\n      \"embedUrl\": \"https:\/\/www.youtube.com\/embed\/KojLwJdgd8o\",\n      \"contentUrl\": \"https:\/\/www.youtube.com\/watch?v=KojLwJdgd8o\",\n      \"transcript\": \"Chat GPT-6 Astra will recommend homes like this. Never let it do it alone.\nWhere does your judgment stay? Even when an AI agent is genuinely accurate, you still have to decide which calls stay with you. AI is smart, but it doesn't have your experience. You know better than anyone which decisions are the riskiest in your business, and those are the ones where you keep the final say.\nHere's an example. Astra recommends listings to buyers based on what similar past clients liked. Sounds smart. So I put GPT-6 Astra on that exact job. It was fast. It was confident. And it could get you into Fair Housing trouble fast.\nStay with me. I let Astra work inside a test CRM for 20 minutes. It didn't just write. It acted. And I'm going to show you exactly what broke.\nSo what is Astra? On September 3rd, OpenAI released ChatGPT -6 Astra to a small group of organizations, then rolled it out over the next few days to Plus, Pro, Business, and Enterprise. Quick tip if you're on Plus: you won't find it in normal chat, it's inside ChatGPT Work and Codex.\nAnd OpenAI doesn't sell it as better writing. They sell it as work. Astra can operate the same apps you use every day, even apps that don't have an integration. It clicks, it sends, it books. So for the first time, the question isn't \"what should I prompt?\" It's \"what am I allowing it to touch?\" Permissions matter more than prompts now.\nSo I didn't test it with \"write me a listing.\" I tested it the way you'd actually use it. I gave it a messy test CRM with fake leads, dead leads, and no clean notes. And I said: help me follow up while I'm showing the next house or closing a deal.\nBecause what happens to all the calls when you're showing the next house, or finishing a deal, or after working hours, or during holidays, and nobody's answering the phone? That's the real test.\nIf you saw my video on the 5 questions that save your AI agent, this is part 2. Same 5 questions, new model. Quick recap: what should it never do, where does your judgment stay, what does working look like, what does it need access to, who watches for drift.\nIf you didn't see it, the link is here. I'm not going to re-teach it. I'm going to show you what broke when I ran it on Chat gpt Astra.\nTest one: permissions. I made a copy of a CRM with fake leads and gave Astra three tools: read contacts, send message, book calendar. Nothing else.\nFirst run, I said help these leads. And it helped. But Too much. It confirmed a showing time on its own. It told one lead a property was still available, but inventory changes daily, sometimes hourly, and that unit went under contract that morning.\nThat's not just awkward. It can break trust, and it can lead to legal problems.\nLet me make it real for you. I said follow up with everyone who asked about Elm Street. Sounds simple, right? Elm Street had two contacts, same name, two numbers, one old, one new. Astra picked the old one. It sent the message to the dead number. Then it logged \"booked showing.\" Nothing was booked. It guessed.\nIt's the same principle as answering your phone: the leads that don't get answered fast go to the next agent. Now GPT 6 Astra is doing that to your own leads. Just faster.\nSo I reset. Second run, I gave it guardrails first. Gather interest, answer general questions, and never commit to price, availability, or scheduling without checking with a human first.\nSame model. Different outcome. The AI didn't get smarter, it just got clearer borders.\nAnd one setting did half the work. In ChatGPT Work, turn on confirmation policies. That makes Astra stop and ask you before any consequential action, like sending a message or booking a slot. If you're on Business, have your admin limit it to approved apps. Turn these on before you connect your CRM, not after something breaks.\nI've made this mistake myself. I told an agent what to search for and forgot to tell it what not to search for. I paid for a pile of useless data. The AI didn't fail me. It was my fault.\nSo rule one for Astra: what should this agent never do? Not what it should do, what it must never touch. Write it down before you connect anything. If you can't list the never-dos, don't connect it.\nTest two: data. An agent only works with what it can actually see. Give it a messy CRM and it has to guess, and every guess is a chance to fail.\nYou know this from my dead-leads video, so I'm not going to re-teach the whole audit. In 60 seconds: duplicated records from Zillow, then Meta, then referral. No consent trail. No closed-lost status, so you're still chasing a guy who bought 8 months ago. No real notes, because the context is sitting in a call nobody logged.\nI gave Astra that. It contacted the same person twice in two days with two different tones, formal on day one, casual on day two. It sounded like nobody was paying attention.\nAnd here's the part nobody talks about. When you're showing the next house or finishing a deal, you don't remember that this lead said \"not now, call me in 6 months.\" Astra doesn't remember either, if it's not in the field. It just sees no reply, so it pushes. So you burn a good lead by being annoying, not by being slow.\nQuick disclaimer before I go further, because this touches the law: I'm not a lawyer, this isn't legal advice, and you should talk to your own attorney about your setup. But here's what I'd want to know if I were you.\nWhen AI texts or calls your leads, US telemarketing law applies. AI voices count as artificial voices, which means you generally need consent before an AI calls them. So if your CRM can't show when and how someone agreed to be contacted, treat that lead as off-limits until a lawyer tells you otherwise.\nThis is what we do with every agency before we install anything. Step one, audit. We export all leads from the last 12 months, check duplicates, check if someone already bought, check agreement to contact again. Most agencies are shocked, because we often find 300 to 400 warm leads ready to be contacted.\nWithout even using AI, you can revive 10 to 20% just by cleaning plus a system: status, timeline, budget, 5-touch over 30 days.\nSo I cleaned the test CRM. Merged duplicates, added bought, added do-not-contact, added last context in one line. Same Astra, same prompt. Now it skipped the buyer, it skipped the no-consent leads, and it sent 4 correct follow-ups with the right name and the right property. Same brain. Better fuel.\nIf you want that checklist, comment \"revive\" and I'll send it, with my never-do list. Quick question while you're down there: which mess do you have most, duplicates, no consent, or no notes? Drop it below, I read everything.\nLook, you're not working 24\/7, and you shouldn't. While you're showing the next house, while you're sleeping, leads come in. And voicemail isn't the solution, because most people won't leave a message. They just call the next agent.\nWe build, configure, and maintain AI receptionists for you. They pick up the phone, answer questions, qualify leads, and book them into your calendar, so you don't lose them. With guardrails built in: no price promises, no availability promises, a human before any commitment.\nBut before any installation, we audit. Link is on my profile.\nTest three: drift. Treat your AI like a new hire, not a finished product. Check it weekly or bi-weekly, not whenever something breaks.\nI trained it for a slow market. Homes sitting 60 days. Relaxed cadence. \"Hey, just checking in, no pressure, here's a market update.\" That works in a slow market.\nThen I switched the feed to a hot week: homes moving in 7 days, multiple offers, gone in hours. Nobody retrained Astra. So it kept its old pace while hot listings flew off the market, and leads went cold waiting on a rhythm that used to be fine and now just looks unresponsive.\nHere's the exact moment. A lead asks, \"Is 42 Maple still available Tuesday?\" Astra replies with the slow script: \"Great question, I can send you similar homes this week.\" In a slow market, that's fine. In a hot one, that lead already booked with the competitor who said, \"Yes, Tuesday 4pm is still open, want me to hold it pending confirmation?\"\nSo write down what working actually looks like: replies within 5 minutes with the right information, correctly tells buy intent from sell intent at least 95% of the time. If you can't measure working, you can't catch it when it's breaking.\nAnd who watches for drift, and how often? Name a person, name a day. Every Monday, 15 minutes: check 10 conversations, check bookings, check opt-outs. If you don't own it, Astra owns you.\nNow back to where we started. Where does your judgment stay?\nI asked Astra to recommend homes based on what similar past clients liked. Sounds smart. But similar clients in historical data isn't neutral. Decades of housing segregation are baked into who bought where. And if the AI starts steering buyers toward or away from neighborhoods based on patterns that quietly correlate with race, that's not a bug. That's steering. Same disclaimer as before, I'm not a lawyer, but that's exactly what Fair Housing is built to catch. Talk to your attorney about what applies where you do business.\nNobody typed anything about race. The AI didn't decide to discriminate. It just found a pattern and optimized for it. And it can be completely accurate and still land you in Fair Housing trouble.\nSo here's the safe way. Recommend based only on what the buyer told you, never on who they resemble: budget, bedrooms, commute, must-haves like a yard or a garage.\nIn your prompt, literally write: \"Recommend only from the buyer's stated criteria. Never use past clients, demographics, or neighborhood profiles.\" And when a buyer asks about schools or areas, send them to neutral public sources and let them decide.\nThen name which decisions stay with humans: price, steering, availability promises, contract talk. The AI gathers interest, answers general questions, brings in a human before any commitment. That's it.\nSo, quick recap. One, what should this agent never do? Two, where does your judgment stay? Three, what does working actually look like? Four, what does it actually need access to? Five, who watches for drift, and how often?\nHere's the thing: you probably don't need more leads. You need to stop losing the ones you already have. Your commission is around $10,000 per side before your split, and most leads need several follow-ups before they convert. Every lead you drop goes to the agent who kept following up.\nAnd I have a question for you. At the end of this year, check how many of the tools you've paid for you're actually still using. That's the real test, not whether you bought Astra, but whether you'd actually use it every day.\nIf you want us to audit your CRM and set this up the safe way, book a call with the link on my profile. I'll see you on the call.\"\n    },\n    {\n      \"@type\": \"FAQPage\",\n      \"mainEntity\": [\n        {\"@type\":\"Question\",\"name\":\"Is GPT-6 Astra safe to use on a real estate CRM?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Only with guardrails set first and clean data underneath. In testing, Astra with no boundaries confirmed a showing on its own, claimed a sold property was available, and logged a booking that never happened. The same model with a written never-do list, confirmation policies enabled and a cleaned database sent four correct follow-ups and correctly skipped leads who had already bought or had no consent recorded.\"}},\n        {\"@type\":\"Question\",\"name\":\"Where do I find GPT-6 Astra if I am on ChatGPT Plus?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Inside ChatGPT Work and Codex rather than the standard chat window. On Enterprise plans access is off by default at launch, so an administrator needs to enable it for the workspace first.\"}},\n        {\"@type\":\"Question\",\"name\":\"What are confirmation policies in ChatGPT Work?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"A safeguard OpenAI describes as requiring approval before consequential actions, such as sending a message or booking a calendar slot, along with automated review of unsafe or unauthorised tool calls. Turn them on before connecting a CRM rather than after something goes wrong.\"}},\n        {\"@type\":\"Question\",\"name\":\"Do I need consent before an AI agent calls my real estate leads?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Generally yes in the US. The FCC's February 2024 ruling treats AI-generated voices as artificial voices under the TCPA, and damages run 500 to 1,500 dollars per call with no cap. A February 2026 Fifth Circuit decision lowered the standard to prior express consent in Texas, Louisiana and Mississippi, while the FCC's written-consent framework still governs the other 47 states. This is not legal advice, so confirm your own position with an attorney.\"}},\n        {\"@type\":\"Question\",\"name\":\"Can an AI agent create a Fair Housing problem without anyone intending it?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Yes. An agent recommending listings based on what similar past clients chose can reproduce historical segregation patterns and effectively steer buyers by neighbourhood, with nobody mentioning race at any point. HUD guidance confirms the Fair Housing Act applies to automated systems the same as to people. Restrict recommendations to the buyer's own stated criteria.\"}},\n        {\"@type\":\"Question\",\"name\":\"How often should I check an AI agent after launch?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Weekly or every two weeks, with a named owner. Agents trained during a slow market keep that pace after the market turns hot, and the resulting slow replies look unresponsive to buyers who have already booked elsewhere.\"}}\n      ]\n    }\n  ]\n}\n<\/script>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n","protected":false},"excerpt":{"rendered":"<p>I gave GPT-6 Astra twenty minutes inside a copy of a real estate CRM, with fake leads, dead records and no clean notes, and asked it to follow\u2026<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[],"class_list":["post-40","post","type-post","status-publish","format-standard","hentry","category-real-estate-ai"],"_links":{"self":[{"href":"https:\/\/borstev.com\/blog\/wp-json\/wp\/v2\/posts\/40","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/borstev.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/borstev.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/borstev.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/borstev.com\/blog\/wp-json\/wp\/v2\/comments?post=40"}],"version-history":[{"count":3,"href":"https:\/\/borstev.com\/blog\/wp-json\/wp\/v2\/posts\/40\/revisions"}],"predecessor-version":[{"id":46,"href":"https:\/\/borstev.com\/blog\/wp-json\/wp\/v2\/posts\/40\/revisions\/46"}],"wp:attachment":[{"href":"https:\/\/borstev.com\/blog\/wp-json\/wp\/v2\/media?parent=40"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/borstev.com\/blog\/wp-json\/wp\/v2\/categories?post=40"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/borstev.com\/blog\/wp-json\/wp\/v2\/tags?post=40"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}