How to Get on AI Podcasts

To understand how to get on AI podcasts, replace the broad "AI expert" pitch with one decision you directly owned. Name the listener, the constraint and the evidence a host can inspect. Then verify that the show is active, uses outside interviews and publishes a legitimate route for nominations or proposals before sending anything.

How to get on AI podcasts with a pitch a producer can assess

AI shows receive claims about changing industries, building the future and knowing what comes next. Those claims give a producer little to test. A specific operating decision gives the interview a usable center.

A machine learning lead might explain why a promising evaluation failed to predict production behavior. Product owners could discuss where human review stayed in the workflow. A founder might describe a choice between model quality, latency and cost, provided the example can be discussed without exposing customer data. Each topic names a conflict and a basis for questioning the guest.

The stronger version can include a failure or unresolved tradeoff. Producers need a truthful conversation, not a victory story.

Pick the AI listener before choosing shows

The word "AI" joins audiences that do different work. Research scientists may care about methods and evidence. AI engineers care about infrastructure, evaluation and operation. Product leaders care about user behavior and workflow. Executives care about ownership, risk and where automation belongs.

Start in the Convokast podcast directory, then use audience overlap to define the role. Read recent episode descriptions and listen to full conversations. Record the questions the host repeats, the technical depth and the kind of evidence guests are asked to provide.

The official pages make these differences visible. Practical AI describes accessible discussions of practical implementations and real-world scenarios. The TWIML AI Podcast has a library spanning machine learning research, models, data and production systems. Latent Space focuses on AI engineering, agents, models, infrastructure and AI for science.

One pitch should not be sent unchanged to all three. An accessible account of a workflow decision may fit Practical AI. A method, dataset or systems discussion may be more relevant to TWIML. An account of agent architecture or evaluation infrastructure may fit Latent Space. Fit still does not prove that any of those pages invites unsolicited proposals.

Verify format, permission and activity separately

An interview archive establishes that selected outside voices appear. A current official application or nomination establishes permission to submit. Recent publishing activity establishes that the program is operating. Check all three conditions before outreach.

The NVIDIA AI Podcast provides the clearest example of permission. Its official page invites visitors to nominate themselves or someone they know. The form supports review, while the publisher keeps control of selection and publication.

Eye on AI publishes long-form interviews and transcripts about research, industry, policy and current events. The presence of interviews establishes format. Its contact navigation does not state that cold guest proposals are open. A similar boundary applies when a site offers reader feedback, sponsorship or newsletter contact without guest-specific wording.

Use the guest-access checklist before outreach. Keep guest applications, nominations, editorial contact, sponsorship and general support as separate route types. If no current route exists, do not scrape a host's personal address to manufacture one.

Build the topic from direct responsibility

A credible topic starts with work the proposed guest performed, supervised or decided. Job title alone is weak evidence. "Founder" does not prove technical depth, and "researcher" does not prove operating experience.

Write the decision in one sentence. State the options considered, the constraint that changed the choice and the result that can be supported. If the result cannot be disclosed, focus on the process or use a public example. Do not combine several customers into one supposed case.

Useful AI interview conflicts include an evaluation that rewarded the wrong behavior, a data limit that changed model choice, a review step that could not be removed, an agent failure that required a simpler workflow or a research result that behaved differently in another setting. These are structures for thinking, not facts to copy into a pitch.

Remove the company name from the proposed episode title. If the subject stops making sense, the idea probably depends on promotion. A vendor can still make a useful guest, but the listener should gain a method, decision rule or caution without becoming a customer.

Prepare evidence before contacting the producer

Create a compact evidence sheet. Separate direct experience, public support, interpretation and restricted material.

Direct experience covers what the speaker personally built, tested, managed or observed in a defined role. Public support includes documentation, research, code, published analysis or approved company material that the host can review. Interpretation is the guest's conclusion and should retain uncertainty. Restricted material includes private customer records, confidential datasets, security details, unpublished research and information governed by employment or client obligations.

Confirm the guest's authority to speak. Expertise does not grant permission to speak for an employer, laboratory, customer or investor. Use first-person language for personal accounts and avoid implied institutional endorsement.

Disclose employment, ownership, investment, vendor and sponsorship interests. A producer can work with a commercial relationship when it is visible. Hidden interests make the episode harder to edit and trust.

Write the submission around the show's editorial job

Open with the listener problem and proposed lesson. Follow with the speaker's direct role and one piece of evidence. Add a short biography only after the producer can see the episode.

Mention a recent episode when it creates a real connection. Explain whether the proposed guest adds a different operating setting, new evidence or a competing interpretation. Generic praise does not show that the sender understands the program.

Follow the requested route and form. A nomination form may ask about the person, organization and subject. Answer every requested field directly. A general contact page should not be relabeled as a guest application. One follow-up can add a material update, while repeated reminders do not repair weak fit.

Before recording, agree on topics that cannot be discussed, the status of any commercial relationship and the source for claims likely to need checking. Prepare plain explanations for technical terms without erasing uncertainty or edge cases.

Once the listener, evidence and official route are settled, write the submission with the podcast pitch generator.