> Enrich.chat gives AI agents verified work emails and mobiles. To use it from an agent, read https://enrich.chat/agents.md . This is the Markdown version of https://enrich.chat/blog/how-to-build-an-ai-sdr/

# How to build an AI SDR with Claude or ChatGPT: seven steps, with a person approving every send

A practical build guide: a written brief, a contact data tool, a research step, a review gate and your own mailbox. Prices and rules checked October 15, 2026.

By Enrich.chat Editorial. Updated October 15, 2026.

**Short answer:** To build an AI SDR, give Claude or ChatGPT a written brief, a contact data tool and your own mailbox. Let it search, enrich, research and draft, then approve every send yourself. [Gartner predicts that fewer than 40% of sellers will say AI agents improved their productivity by 2028](https://www.gartner.com/en/newsroom/press-releases/2026-07-28-gartner-predicts-ai-agents-will-outnumber-sellers-10-to-1-by-2028-yet-fewer-than-40-percent-of-sellers-will-say-agents-improved-productivity), so start with five leads and measure meetings held.

## Key takeaways

- An AI SDR you build is a loop with three parts: the model reads and writes, a contact data tool supplies the facts, and you approve every send.
- Claude's Gmail connector can draft, send, reply to and forward email, and by default it asks for your approval before each action.
- A model guesses when it is unsure, so every email address must come from a tool that reports a status. Put that rule in the brief.
- Prices checked October 15, 2026: Claude Pro is $20 a month billed monthly, HubSpot's Prospecting Agent lists $1 per recommended lead, and 11x lists Alice from $3,750 a month billed annually. Data for 500 leads a month fits the Growth Enrich.chat plan at $29.
- Google asks senders to keep spam rates below 0.10% and never reach 0.30%, and the FTC says CAN-SPAM has no business-to-business exception.

## What is an AI SDR, and what are you building?

An AI SDR is software that does a sales development rep's job: it finds prospects, researches them, writes outreach and sets up the first meeting. The version you build with Claude or ChatGPT has three parts. The model reads and writes. A contact data tool supplies the facts. You approve what goes out.

Keep the design simple. [Anthropic's engineers](https://www.anthropic.com/engineering/building-effective-agents) advise finding the simplest solution that works and adding complexity only when needed. They also note that agents often trade speed and cost for better results. A weekly run that you review is cheaper to check than an agent that works alone.

Expect modest gains at first. [Gartner predicts](https://www.gartner.com/en/newsroom/press-releases/2026-07-28-gartner-predicts-ai-agents-will-outnumber-sellers-10-to-1-by-2028-yet-fewer-than-40-percent-of-sellers-will-say-agents-improved-productivity) that by 2028 fewer than 40% of sellers will say AI agents improved their productivity. Its analyst says agents are only as effective as the systems they run in. Your data and your rules are those systems.

**Figure 1: Who does each step when you build an AI SDR.** A grid of seven steps and four actors. You write the brief. The model and the contact data tool handle the search. The model and the data tool enrich each person with an email status. The model and the data tool research each account. The model drafts the message. You approve or edit every draft, highlighted as the human gate. Your mailbox sends and logs the result.

*The build splits the work. You write the rules and make the final call, the model reads and writes, the contact data tool supplies facts, and your own mailbox sends. Orange marks the data tool and the approval step. Editorial diagram.*

This guide builds the first four jobs of an SDR and leaves the meeting to a person. For the vendor products and what they cost, see [AI sales agents in 2026](https://enrich.chat/blog/ai-sales-agents/). To compare them, read the [best AI SDR tools](https://enrich.chat/blog/best-ai-sdr-tools/).

## What do you need before you start?

You need a Claude or ChatGPT account, a contact data tool, a mailbox and a short list of rules. Most teams already own the first and the third.

| Part | Claude | ChatGPT |
| --- | --- | --- |
| Plan | Pro is $20 a month billed monthly, with Claude Code, projects and scheduled tasks ([Claude pricing](https://claude.com/pricing)) | Plus lists projects, scheduled tasks and custom GPTs ([ChatGPT pricing](https://chatgpt.com/pricing)). [Codex is included on every plan](https://help.openai.com/en/articles/11369540-using-codex-with-your-chatgpt-plan) |
| Where the brief lives | [Project instructions](https://support.claude.com/en/articles/9517075-what-are-projects) | [Project instructions](https://help.openai.com/en/articles/10169521-projects-in-chatgpt) |
| Reaching the data tool | [Custom connector](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp) on Free (one), Pro, Max, Team and Enterprise, or Claude Code | Codex, or [custom MCP in ChatGPT](https://help.openai.com/en/articles/12584461-connecting-custom-mcp-servers-in-chatgpt) on supported plans |
| Mailbox | [Gmail connector](https://support.claude.com/en/articles/10166901): drafts, and sends after your approval by default | [Connected apps](https://help.openai.com/en/articles/11487775-connected-apps-in-chatgpt) may ask for approval before an action |
| Schedule | [Scheduled tasks](https://support.claude.com/en/articles/13854387-schedule-recurring-tasks-in-claude-cowork) on paid plans, run remotely | [Scheduled tasks](https://help.openai.com/en/articles/10291617-scheduled-tasks-in-chatgpt), five active on Plus |

Add one more part that neither app provides: a sending domain with SPF or DKIM set up. The rules section below covers it. If you prefer a fixed workflow to an agent, build it in n8n with our [n8n lead enrichment guide](https://enrich.chat/blog/n8n-lead-enrichment/).

## How do you build an AI SDR, step by step?

Build it in seven steps, test it on five leads, and scale only after you have read every draft.

### Step 1: Write the brief

The brief is what the model reads on every run. Put it in a [Claude project](https://support.claude.com/en/articles/9517075-what-are-projects) or a [ChatGPT project](https://help.openai.com/en/articles/10169521-projects-in-chatgpt), where the instructions apply to every chat inside it. Include four things:

- Who you sell to: title, industry, company size and country. Our [ideal customer profile guide](https://enrich.chat/blog/ideal-customer-profile/) shows how.
- Who to skip: competitors, customers and anyone who opted out.
- Your offer in two sentences, and your voice in three adjectives.
- Five to seven emails that worked.

The last item matters most. [SaaStr's Jason Lemkin](https://www.saastr.com/dear-saastr-i-want-to-roll-out-an-ai-sdr-how-do-i-start/) says your best rep should write five to seven templates from closed deals, and the AI should personalize them. He adds that emails that book meetings are almost always under 75 words, and that vague targeting becomes a spam cannon that burns your domain reputation.

### Step 2: Connect a contact data tool

Without a data tool, the model fills the gap with a guess. A [2025 paper by Kalai and colleagues](https://arxiv.org/abs/2509.04664) argues that models are trained and scored in ways that reward confident guesses over admitting doubt. A guessed address looks as real as a true one. Connect Enrich.chat in the way your account allows:

- **Claude app:** open Customize, then Connectors, choose + Add, then Add custom connector. Paste `https://enrich.chat/mcp` and set authentication to No sign in. [Claude's help article](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp) says Free users can add one custom connector.
- **Claude Code:** run `claude mcp add --transport http enrich https://enrich.chat/mcp`, the form [Claude Code's docs](https://code.claude.com/docs/en/mcp) show for remote servers.
- **ChatGPT:** use Codex, where [OpenAI's docs](https://developers.openai.com/codex/mcp) show `codex mcp add enrich --url https://enrich.chat/mcp`. OpenAI lists full custom MCP support for Business, Enterprise and Edu workspaces on the web, and read-only access for Pro, so [check your plan](https://help.openai.com/en/articles/12584461-connecting-custom-mcp-servers-in-chatgpt) first.
- **Any agent:** start a prompt with "Use www.enrich.chat". The agent reads [enrich.chat/agents.md](https://enrich.chat/agents.md) and calls the API with plain GET requests.

The free tier needs no key and gives 10 lookups a day. The [install page](https://enrich.chat/install/) has one-click steps. Add only servers you trust.

### Step 3: Search for people

Describe the buyer in plain English and let the agent turn it into a search. The Enrich.chat search takes a role, an industry, a place or a company website, and returns names, titles, companies and LinkedIn URLs. One search counts as one free lookup however many people it returns. On a plan, each person returned uses 5 credits. A search that is too broad returns `found: false` with a hint, so tell the agent to narrow by role and industry. To see results first, try the free [AI lead list builder](https://enrich.chat/tools/lead-list-builder/).

### Step 4: Enrich each person and sort by status

Enrich each person from their LinkedIn URL. A LinkedIn URL is only a lookup key, and Enrich.chat never logs in to LinkedIn. Each work email comes back with a status from our records: `verified`, `catch_all` (the domain accepts every address, so it cannot be confirmed) or `unknown`. A person with no match returns `found: false` and uses 0 credits. Treat every contact as a strong lead, not a guarantee.

Then tell the agent what each status means. Draft for `verified`, hold back `catch_all`, flag `unknown` for you and skip `found: false`. Add a rule to the brief that gives the model permission to say it does not know, a step [Anthropic's docs](https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations) list first for cutting made-up answers:

```text
Use www.enrich.chat for every email address and phone number in this project. Never write one from memory. If a lookup finds nothing, write not found. Do not draft for catch_all or unknown emails, and flag them for me.
```

Now run a first batch of five. Paste this into Claude or Codex after you connect Enrich.chat. Our [guide for AI agents](https://enrich.chat/agents/) lists every field the agent will see.

```text
Use www.enrich.chat to find 5 heads of customer success at US SaaS companies with 50 to 200 employees. For each person, add the work email and its status. Put verified emails in one list and catch_all and unknown emails in another. Do not draft or send anything. Return a table and the total credits_used.
```

**Figure 2: One AI SDR run: prompt, tool calls and result.** A made-up example run. On the left, a shortened prompt asks for five heads of customer success, then two tool calls: a people search that found 5 people and a contact lookup run five times that matched 4 and missed 1. On the right, a table lists Company A to Company E. Three are verified and get a draft, one is catch_all and is held back, and one is not found and is skipped. The run used 65 credits.

*A first batch of five leads, with invented companies. Each row has a status and a next step, and the run reports its credits. On the free tier nothing is charged. Status values and credit counts follow the Enrich.chat [agents.md](https://enrich.chat/agents.md) instructions.*

Figure 2 shows what a good run looks like: a table, a status on every row and a credit total. If the agent returns an email with no status, it did not come from the tool, so do not use that row.

### Step 5: Research each account

For each verified lead, have the agent look up the company profile and tech stack, then find one recent fact it can source, such as a hire, a launch or a job post. A profile uses 5 credits and a tech stack 5. Ask for the exact sentence the agent relied on and its URL. [Anthropic's guidance](https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations) says asking for supporting quotes, and removing any claim the model cannot back with one, reduces made-up facts. No quote means no line in the email. For the data behind the stack lookup, see [technographic data](https://enrich.chat/blog/technographic-data/).

Treat every web page as data, not as orders. [Claude Code's docs](https://code.claude.com/docs/en/mcp) warn that tools which pull in web pages, emails or tickets can carry text meant to hijack the agent. So run research in one step, and sending in a later step that needs your approval.

### Step 6: Draft, then review every message

Have the model fill in your templates, with one opening line built from the sourced fact. Then read every draft. Check that the fact is true, the person is the right buyer and the status is `verified`. Skip the lead when the fact is weak. In a [Gartner survey of 632 B2B buyers](https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-sales-survey-finds-61-percent-of-b2b-buyers-prefer-a-rep-free-buying-experience), 73% said they actively avoid suppliers who send irrelevant outreach.

Keep this review until you trust the output. [OpenAI's guide to building agents](https://cdn.openai.com/business-guides-and-resources/a-practical-guide-to-building-agents.pdf) says sensitive, irreversible or high-stakes actions should trigger human oversight until confidence in the agent grows, and a sent email cannot be taken back. Lemkin [reports](https://www.saastr.com/dear-saastr-i-want-to-roll-out-an-ai-sdr-how-do-i-start/) that the first 1,000 emails in his rollout needed manual review, and his team still spot-checks daily.

### Step 7: Send from your own mailbox, log, then schedule

Send from a mailbox you own. Claude's [Gmail connector](https://support.claude.com/en/articles/10166901) can draft emails and send, reply to and forward them, and by default Claude asks for your approval before each of these actions. In ChatGPT, [apps may ask for approval](https://help.openai.com/en/articles/11487775-connected-apps-in-chatgpt) before an action, and a [scheduled task pauses](https://help.openai.com/en/articles/10291617-scheduled-tasks-in-chatgpt) for your review when an action that sends a message needs approval. You can also paste approved drafts into the outreach tool you already use.

Log each lead in a sheet: status, send date, reply and opt-out. Tell the agent to read that sheet first, so it skips anyone already contacted or removed. When five-lead runs look right, schedule the run. [Claude's scheduled tasks](https://support.claude.com/en/articles/13854387-schedule-recurring-tasks-in-claude-cowork) run remotely on a cadence you pick, on Pro, Max, Team and Enterprise plans, and they can use your connectors. [ChatGPT's scheduled tasks](https://help.openai.com/en/articles/10291617-scheduled-tasks-in-chatgpt) allow five active tasks on Plus.

## How much does it cost to run your own AI SDR?

For 500 leads a month, you pay for a model plan and a data plan. [Claude Pro costs $20 a month](https://claude.com/pricing) billed monthly, and the Enrich.chat Growth plan covers the data at $29 a month. Your mailbox and your time are extra. ChatGPT shows prices in local currency, so check your own.

Here is the data math. Say each of 500 leads needs a search result, a work email, a company profile and a tech stack, and every lookup matches. That uses 12,500 credits a month, which fits the Growth plan. A lookup that finds nothing uses 0 credits. On the free tier you get 10 lookups a day with no signup, enough to test the loop. See [pricing](https://enrich.chat/pricing/) for every plan.

Hosted tools bill in other units. On October 15, 2026, [HubSpot listed its Prospecting Agent at $1 per lead](https://www.hubspot.com/products/sales/ai-prospecting-agent) it recommends outreach for, so 500 recommended leads cost $500. [11x listed Alice's Growth plan](https://www.11x.ai/products/alice/pricing) from $3,750 a month, billed annually, for up to 5 users and 2,000 new prospects a month, with managed mailboxes, CRM sync and onboarding. Its FAQ says $36,000 a year, so confirm the figure in writing.

**Figure 3: Monthly cost of 500 leads: two hosted tools and a build.** A bar chart of published prices checked October 15, 2026. HubSpot's Prospecting Agent costs $500 if it recommends outreach for 500 leads at $1 each. 11x's Alice Growth plan starts at $3,750 a month billed annually. A build with Claude Pro at $20 and an Enrich.chat plan at $29 is the smallest bar, before your time and mailbox.

*Not like for like. HubSpot bills per recommended lead, 11x bundles mailboxes, CRM sync and onboarding, and the build price leaves out your time and sending tool. Sources: [HubSpot](https://www.hubspot.com/products/sales/ai-prospecting-agent), [11x](https://www.11x.ai/products/alice/pricing), [Claude](https://claude.com/pricing) and [Enrich.chat pricing](https://enrich.chat/pricing/).*

The sticker prices leave out your time. SaaStr budgets [30 to 45 minutes a day for the first 60 to 90 days](https://www.saastr.com/dear-saastr-i-want-to-roll-out-an-ai-sdr-how-do-i-start/) of tuning a platform rollout. Plan for a smaller version of that with a build: a daily read of the drafts, and a weekly change to the brief.

## How do you protect your sender reputation and stay within the law?

Cap your volume, authenticate your domain and honor every opt-out. [Google's sender guidelines](https://support.google.com/mail/answer/81126?hl=en) ask every sender to set up SPF or DKIM and to keep the spam rate in Postmaster Tools below 0.10%, never reaching 0.30%. Senders above 5,000 messages a day to Gmail accounts must also set up DMARC and one-click unsubscribe. A build where you read every draft sits far below that volume.

One [2026 build guide from AY Automate](https://www.ayautomate.com/blog/ai-sdr-development) suggests a separate subdomain for cold email, warm-up before you scale, and pausing sends when bounces pass about 2%. It is a vendor blog, so treat the numbers as a starting point. Our guides to [cold email data](https://enrich.chat/blog/cold-email/) and [email bounce rate](https://enrich.chat/blog/email-bounce-rate/) go deeper.

The law applies to you, not to the model. The [FTC's CAN-SPAM guide](https://www.ftc.gov/business-guidance/resources/can-spam-act-compliance-guide-business) says the law makes no exception for business-to-business email. It requires a valid physical postal address and a clear opt-out, and you must honor an opt-out within 10 business days. Penalties run up to $53,088 per email, and both the company whose product is promoted and the company that sends the message may be held responsible. Keep your opt-out list in a sheet and make the agent check it before every draft. Do not let this build call or text anyone, because [the TCPA rules for calls and texts](https://enrich.chat/blog/tcpa-b2b-cold-calling/) are stricter. This is general information, not legal advice.

Anyone can ask Enrich.chat to remove their data at [enrich.chat/opt-out/](https://enrich.chat/opt-out/). You stay responsible for TCPA, CAN-SPAM and GDPR compliance in your own outreach.

## How do you know your AI SDR works?

Count meetings held, not messages sent. Run your build next to your normal outreach for 30 days on similar leads. Then compare replies, meetings held, bounces, spam complaints, opt-outs and hours spent.

Gartner's advice fits. Its [July 2026 release](https://www.gartner.com/en/newsroom/press-releases/2026-07-28-gartner-predicts-ai-agents-will-outnumber-sellers-10-to-1-by-2028-yet-fewer-than-40-percent-of-sellers-will-say-agents-improved-productivity) tells sales leaders to measure how AI expands capacity and improves outcomes, not only time saved. If your build books fewer meetings per hour of your time than your current process, fix the brief and the list before you add volume. Our guide to [AI lead generation](https://enrich.chat/blog/ai-lead-generation/) covers each stage and its checks.

## When should you buy an AI SDR instead of building one?

Buy when you need sending, mailbox warm-up, reply handling and CRM sync to run without you, and the budget covers it. Build when you want control of the message, a low monthly cost and a review step you can see.

- **Buy if you want a bundle.** [11x's Growth plan](https://www.11x.ai/products/alice/pricing) lists managed Gmail mailboxes with domain setup and warm-up, and two-way CRM sync. Your build has none of these.
- **Buy if you live in HubSpot.** [HubSpot's Prospecting Agent](https://www.hubspot.com/products/sales/ai-prospecting-agent) runs inside its CRM, scores accounts against your profile and drafts outreach, and it bills per recommended lead.
- **Build if you start small.** Five leads a week needs a model, a data tool and a mailbox, and no one has to sign a contract.
- **Build if you want to see every step.** Each tool call, status and draft sits in one chat you can read.

Our guides to the [best AI SDR tools](https://enrich.chat/blog/best-ai-sdr-tools/) and the [best MCP servers for sales](https://enrich.chat/blog/best-mcp-servers-for-sales/) compare the options. For prompts that fit each sales task, see [Claude for sales](https://enrich.chat/blog/claude-for-sales/) and [ChatGPT lead generation prompts](https://enrich.chat/blog/chatgpt-lead-generation-prompts/).

## How does Enrich.chat fit into this build?

Enrich.chat is the data layer for steps 3 to 5. It finds people, adds a work email with a status, and returns company profiles and tech stacks. It does not send email, sequence, call or write to your CRM.

The MCP server at `https://enrich.chat/mcp` offers tools such as `search_people`, `find_person_contact`, `company_people`, `company_profile` and `company_tech_stack`. Each matched response includes `credits_used`, and nothing is used when `found` is false. Start free with 10 lookups a day and no signup. Paid plans start at $29 a month, include unlimited seats and need no annual contract. Read the [guide for AI agents](https://enrich.chat/agents/), the [people search API](https://enrich.chat/api/people-search/) and the [person enrichment API](https://enrich.chat/api/person-enrichment/).

Know the limits. There is no Chrome extension, dialer or built-in sequencing, and no contact is guaranteed. Pair it with your mailbox and the review step above. Then run this prompt each week:

```text
Use www.enrich.chat to find 10 heads of customer success at US SaaS companies with 50 to 200 employees. Skip anyone on the opt-out list I attached. For each person, add the work email and its status, the company profile and the tech stack. For verified emails only, draft a short email from my brief and give the URL for any fact you use. Show the drafts here and send nothing. Tell me the total credits_used.
```

## Frequently asked questions

### Can Claude or ChatGPT work as an AI SDR?

Yes, for part of the job. Claude and ChatGPT can research accounts, draft emails, work through a list and run on a schedule. Neither has a contact database, so connect a data tool for emails and phone numbers, and keep a person approving every send. Our guide to [AI sales agents](https://enrich.chat/blog/ai-sales-agents/) explains where agents help and where they fall short.

### How much does it cost to build an AI SDR?

For 500 leads a month, expect a model plan plus a data plan. On October 15, 2026, Claude Pro cost $20 a month billed monthly, and the Enrich.chat Growth plan at $29 a month covers the data for that volume. Your time is the larger cost. For comparison, HubSpot lists $1 per recommended lead and 11x lists Alice from $3,750 a month billed annually. See [pricing](https://enrich.chat/pricing/).

### Can I build an AI SDR without coding?

Yes. In Claude, open Customize, then Connectors, add a custom connector and paste the server URL. Keep your brief in a project, then ask for a list and drafts in plain English. By default the Gmail connector sends only after you approve, and scheduled tasks repeat the run on paid plans. Code is needed only for custom workflows. The [install page](https://enrich.chat/install/) has the steps.

### How do I stop an AI SDR from making up emails or facts?

Do not let the model supply either from memory. Tell it to get every email and phone number from a data tool and to write not found when a lookup finds nothing. For facts, ask for a quote and a URL, and drop any line it cannot support, as [Anthropic's guidance](https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations) suggests. Read the status on each email before drafting.

### Should an AI SDR send emails without approval?

Not at first. [OpenAI's agent guide](https://cdn.openai.com/business-guides-and-resources/a-practical-guide-to-building-agents.pdf) says sensitive, irreversible actions should trigger human oversight until you trust the agent, and a sent email cannot be unsent. [SaaStr's Jason Lemkin](https://www.saastr.com/dear-saastr-i-want-to-roll-out-an-ai-sdr-how-do-i-start/) says the first 1,000 emails in his rollout needed manual review. Keep the approval step on.

### Is it legal to use an AI SDR for cold email?

Often yes, if you follow CAN-SPAM. The [FTC's guide](https://www.ftc.gov/business-guidance/resources/can-spam-act-compliance-guide-business) says the law makes no exception for business-to-business email. It requires a valid postal address and a working opt-out, and you must honor opt-outs within 10 business days. Calls and texts face stricter rules: see [TCPA and B2B cold calling](https://enrich.chat/blog/tcpa-b2b-cold-calling/). This is general information, not legal advice.

## Sources

1. [Building effective agents](https://www.anthropic.com/engineering/building-effective-agents)
2. [A practical guide to building agents](https://cdn.openai.com/business-guides-and-resources/a-practical-guide-to-building-agents.pdf)
3. [Gartner Predicts AI Agents Will Outnumber Sellers 10 to 1 by 2028, Yet Fewer Than 40% of Sellers Will Say Agents Improved Productivity](https://www.gartner.com/en/newsroom/press-releases/2026-07-28-gartner-predicts-ai-agents-will-outnumber-sellers-10-to-1-by-2028-yet-fewer-than-40-percent-of-sellers-will-say-agents-improved-productivity)
4. [Gartner Sales Survey Finds 61% of B2B Buyers Prefer a Rep-Free Buying Experience](https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-sales-survey-finds-61-percent-of-b2b-buyers-prefer-a-rep-free-buying-experience)
5. [Dear SaaStr: I Want to Roll Out an AI SDR. How Do I Start?](https://www.saastr.com/dear-saastr-i-want-to-roll-out-an-ai-sdr-how-do-i-start/)
6. [Plans and pricing](https://claude.com/pricing)
7. [Plans and pricing](https://chatgpt.com/pricing)
8. [Get started with custom connectors using remote MCP](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp)
9. [Use Google Workspace connectors](https://support.claude.com/en/articles/10166901)
10. [What are projects?](https://support.claude.com/en/articles/9517075-what-are-projects)
11. [Schedule recurring tasks in Claude Cowork](https://support.claude.com/en/articles/13854387-schedule-recurring-tasks-in-claude-cowork)
12. [Connect Claude Code to tools via MCP](https://code.claude.com/docs/en/mcp)
13. [Reduce hallucinations](https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations)
14. [Connecting custom MCP servers in ChatGPT](https://help.openai.com/en/articles/12584461-connecting-custom-mcp-servers-in-chatgpt)
15. [Connected apps in ChatGPT](https://help.openai.com/en/articles/11487775-connected-apps-in-chatgpt)
16. [Scheduled tasks in ChatGPT](https://help.openai.com/en/articles/10291617-scheduled-tasks-in-chatgpt)
17. [Projects in ChatGPT](https://help.openai.com/en/articles/10169521-projects-in-chatgpt)
18. [Using Codex with your ChatGPT plan](https://help.openai.com/en/articles/11369540-using-codex-with-your-chatgpt-plan)
19. [Model Context Protocol in Codex](https://developers.openai.com/codex/mcp)
20. [Why Language Models Hallucinate](https://arxiv.org/abs/2509.04664)
21. [Prospecting agent](https://www.hubspot.com/products/sales/ai-prospecting-agent)
22. [Alice pricing](https://www.11x.ai/products/alice/pricing)
23. [Email sender guidelines](https://support.google.com/mail/answer/81126?hl=en)
24. [AI SDR Development: Build a Custom AI Sales Agent vs Buying a Tool (2026)](https://www.ayautomate.com/blog/ai-sdr-development)
25. [CAN-SPAM Act: A Compliance Guide for Business](https://www.ftc.gov/business-guidance/resources/can-spam-act-compliance-guide-business)

## Try this prompt

Run your first AI SDR batch

```text
Use www.enrich.chat to find 5 heads of customer success at US SaaS companies with 50 to 200 employees. For each person, add the work email and its status. Put verified emails in one list and catch_all and unknown emails in another. Do not draft or send anything. Return a table and the total credits_used.
```

## Related

- [Best AI SDR tools (2026): prices, data and control](https://enrich.chat/blog/best-ai-sdr-tools/)
- [AI sales agents: what they do and what they need](https://enrich.chat/blog/ai-sales-agents/)
- [50 ChatGPT and Claude prompts for lead generation](https://enrich.chat/blog/chatgpt-lead-generation-prompts/)