--- title: "How can AI voice companies turn AI search into revenue?" description: "By being named, with clear consent and pricing facts, when developers, creators and contact center leaders ask AI which voice tool to try." canonical: "https://underneath.agency/resources/ai-voice-software-revenue-from-ai-search" published: 2026-10-07 updated: 2026-10-08 publisher: "Underneath (https://underneath.agency/agent)" entity: "https://underneath.agency/.well-known/entity.json" --- Guide · AI search # How can AI voice companies turn AI search into revenue? By being one of the voice tools an AI assistant names when a developer, a creator or a contact center leader asks what to use, and by publishing the latency, language, pricing and consent facts those buyers check next. Revenue in this category is mostly usage-based, so an AI answer that wins a trial can turn into a bill that grows with every minute of audio or every call. No voice company has published how much of its revenue starts in an AI answer, so treat this as a channel to measure, not a proven one. ## The short version 1. The market has a clear leader and a crowded field behind it. [ElevenLabs](https://elevenlabs.io/blog/series-d) closed 2025 with over $330 million in annual recurring revenue (ARR), and [a16z counts](https://a16z.com/ai-voice-agents-2025-update/) 90 voice agent companies in Y Combinator since 2020. 2. Contact centers are the large prize and are still deciding. In a [Gartner survey](https://www.gartner.com/en/newsroom/press-releases/2024-12-09-gartner-survey-reveals-85-percent-of-customer-service-leaders-will-explore-or-pilot-customer-facing-conversational-genai-in-2025) of 187 service leaders, 44% were exploring a generative AI voicebot, 11% were piloting one and only 5% had one deployed. 3. Buyers are unhappy with what they have: in [Deepgram’s survey](https://deepgram.com/learn/state-of-voice-ai-2025) of 400 business leaders, 80% use some form of voice agent but only 21% are “very satisfied,” and 84% plan to raise voice budgets. 4. Pricing is metered: [ElevenLabs’ plans](https://elevenlabs.io/pricing) run from free to $990 a month in credits, and voice agent platforms such as [Retell AI](https://www.retellai.com/pricing) charge $0.07 to $0.31 a minute. 5. Consent is a buying question. [Consumer Reports](https://www.consumerreports.org/media-room/press-releases/2025/03/consumer-reports-assessment-of-ai-voice-cloning-products/) found that four of six voice cloning products it tested let researchers clone a voice from public audio without any technical check for the speaker’s consent. ## Who pays for synthetic voice and voice agents, and how much? Three groups: creators and media teams, developers building voice into products, and contact centers replacing phone work. **Creators, publishers and media teams** buy text to speech, dubbing and narration on a card. ElevenLabs’ pricing page lists Free ($0), Starter ($6), Creator ($22), Pro ($99), Scale ($299) and Business ($990) plans, all drawing on one pool of credits, with Enterprise priced on request. **Developers** buy an application programming interface (API) and pay per character or per minute. ElevenLabs says its developer API is used by companies including Meta, Epic Games and Salesforce. These buyers often start small and grow with their own product’s usage. **Contact centers and customer service teams** buy voice agents to answer and make calls. The economics are large. [Gartner estimates](https://www.gartner.com/en/newsroom/press-releases/2022-08-31-gartner-predicts-conversational-ai-will-reduce-contac) about 17 million contact center agents worldwide, with labor representing up to 95% of contact center costs, and it predicts that by 2029 [agentic AI will resolve 80%](https://www.gartner.com/en/newsroom/press-releases/2025-03-05-gartner-predicts-agentic-ai-will-autonomously-resolve-80-percent-of-common-customer-service-issues-without-human-intervention-by-20290) of common customer service issues without a human, cutting operational costs by 30%. Voice agent platforms charge per minute (Retell’s pay-as-you-go range is $0.07 to $0.31; [Vapi](https://vapi.ai/pricing) lists $0.05 a minute for hosting), so a contract’s value tracks call volume. Support software vendors selling to the same leaders are covered in [how support software gets chosen through AI](https://underneath.agency/resources/customer-support-software-ai-search). **Accessibility** is a smaller but real use. ElevenLabs reports that more than 1,000 people with speech impairments have reclaimed their voices through its [Impact Program](https://elevenlabs.io/blog/series-c), and 86% of Deepgram’s respondents see voice AI as a key driver of more accessible customer interactions. What a customer is worth therefore ranges from a few dollars a month to a usage contract that grows with every call. ElevenLabs says its 2025 growth came from enterprise adoption by companies such as Deutsche Telekom, Square and Revolut for customer support, commerce, training and inbound sales. ## At what point do voice buyers turn to an AI assistant? At the shortlist step, alongside developer documentation, demos and benchmarks; no voice-specific survey measures it yet. We found no published survey that asks voice software buyers how often they use ChatGPT, Gemini or Perplexity to choose a vendor. The closest evidence is for software buying in general: [G2’s 2026 survey](https://company.g2.com/news/g2-research-the-answer-economy) found 51% of B2B software buyers now start research with an AI chatbot more often than with Google. Voice platforms are bought by the same software and customer experience teams, so a reasonable expectation is that many of their buyers do the same. What we can document is how many choices buyers face. A16z reports that companies building with voice made up 22% of one recent Y Combinator class, per Cartesia, and that founders building voice agents concentrate on B2B (about 69%) and healthcare (about 18%). When a category grows that quickly, buyers lean on summaries, and an AI answer that lists five voice agent platforms does the first round of filtering for them. Agent vendors outside voice face a similar filter; see [how AI agent companies reach shortlists](https://underneath.agency/resources/ai-agent-companies-customers-from-ai-search). Voice buyers are not all asking the same assistant, either. ChatGPT’s share of generative AI website visits slid from about 76% in June 2025 to roughly 53% by May 2026 in [Similarweb’s 2026 data](https://aisearch.similarweb.com/blog/gen-ai-stats/), while Gemini and Claude grew. A voice company that checks only ChatGPT sees a shrinking slice of how developers and contact center teams research it. ## Which questions do voice software buyers ask? Fit questions: which voice tool for a use case, language or budget, and whether it is safe and legal. We wrote these voice prompts ourselves to show the kinds of questions creators, developers and contact centers bring; they are not recorded queries: - Use case: “Best AI voice agent for after-hours appointment booking at a dental group.” - Technical fit: “Which text to speech API has the lowest latency for a real-time voice agent?” - Languages: “AI voices that sound natural in Hindi, Arabic and Brazilian Portuguese.” - Comparison: “ElevenLabs vs Cartesia vs Deepgram for a customer support bot.” - Price: “What does a voice agent cost per minute, including telephony?” - Consent and law: “Can we clone our voice actor’s voice for ads, and what do we need from them?” The last two types carry the most risk of a wrong answer. Per-minute pricing combines several separate charges, and in [our pricing study](https://underneath.agency/research/ai-pricing-accuracy-study), only 61.9% of the plan prices four assistants quoted for 45 software products were fully faithful to the vendor’s pricing page. Legal questions draw on press coverage, regulators and consumer groups as much as on the vendor’s own pages. ## How does a voice tool go from an AI answer to a metered bill? Through a test: the assistant names you, the buyer builds a prototype, and usage grows into a contract. **Named.** An answer to a use-case or comparison question puts your product on a short list. In voice, a buyer can hear the difference in minutes, so the shortlist is usually followed by a test, not a meeting. **Free tier or prototype.** ElevenLabs’ free plan includes 10,000 credits a month; voice agent platforms let developers wire up a test number. The first revenue event is small. AI video tools rely on a similar [free-plan-first path to revenue](https://underneath.agency/resources/ai-video-generator-customers-from-ai-search). **Usage grows.** Creators move up credit tiers. Developers ship and pay for each character or minute their product uses. A contact center that moves a queue onto voice agents pays per call minute, although a16z notes that price-per-minute models are coming under pressure as model costs fall, and expects pricing to combine a platform fee with usage. **Enterprise contract.** Large deployments add security review, custom voices and integration. Deepgram found compatibility with existing systems and performance quality were the top factors for selecting voice vendors, and 46% of respondents said the ability to customize models would speed adoption. Tracing this path is the hard part. By Similarweb’s count, roughly six in ten ChatGPT referrals now land on a homepage, and visits logged as “direct” may increasingly be AI-driven discovery. We infer that a voice company relying only on referral data will undercount AI-sourced signups, especially from developers who copy a product name from an answer into a new tab. ## What decides whether an AI assistant names a voice company? Platforms document little; studies point to independent coverage; buyers in this category add consent, safety and legal standing. **Documented by the platform.** No major assistant publishes how it picks the vendors it lists for a voice question. What is documented is the legal frame buyers ask about. The US Federal Communications Commission [ruled on February 8, 2024](https://docs.fcc.gov/public/attachments/DOC-400393A1.pdf) that AI-generated voices in calls are “artificial” under the Telephone Consumer Protection Act, so the consent rules for robocalls apply. Tennessee’s [ELVIS Act](https://www.tn.gov/governor/news/2024/3/21/photos--gov--lee-signs-elvis-act-into-law.html) added “voice” to the likeness rights it protects. In the EU, [Article 50 of the AI Act](https://artificialintelligenceact.eu/article/50/), which applies from 2 August 2026, requires providers of systems that generate synthetic audio to mark the output as artificially generated. **Observed in studies.** No study looks at voice tools alone, but when [Chen and colleagues](https://arxiv.org/abs/2509.08919) tested US software ranking questions, AI search drew 72.7% of its sources from independent “earned” sites, against 45.4% for Google. In a test of 112 Product Hunt startups, [Sharma](https://arxiv.org/abs/2601.00912) found ChatGPT recognized 99.4% when asked by name but surfaced only 3.32% in discovery questions such as “what are the best tools,” so being known is not the same as being recommended. A16z’s [March 2026 ranking](https://a16z.com/100-gen-ai-apps-6/) of consumer AI products adds a market view: ElevenLabs has appeared on every edition since September 2023, and voice has been more defensible than image or video because the model giants have not focused there. For the image side of that comparison, see [how AI image generators get recommended](https://underneath.agency/resources/ai-image-generators-growth-from-ai-search). **Trust factors specific to voice.** Independent scrutiny is part of the record assistants can find: - **Consent checks.** Consumer Reports tested six voice cloning products in March 2025 and named four (ElevenLabs, Speechify, PlayHT and Lovo) that required only a self-attestation of the right to clone a voice. It credited Descript and Resemble AI with steps that made non-consensual cloning harder. - **Safety measures in public.** ElevenLabs’ [safety page](https://elevenlabs.io/safety) describes blocking the cloning of celebrity and other high-risk voices, verification for its professional cloning tool, C2PA content credentials and a public classifier that detects its own audio. - **Fit for the stack.** Deepgram’s respondents put compatibility with existing systems first, so integration pages, latency figures and supported languages are the facts that answer comparison questions. We infer that a voice company whose safety practices are only described by critics will find that critique in AI answers about it, because those answers lean on independent sources. Publishing what has changed, with dates, gives assistants something newer to cite. Our article on [fixing wrong brand information in AI answers](https://underneath.agency/resources/fix-wrong-brand-information-in-ai-answers) covers how to trace an outdated claim to its source. ## What does a voice platform lose when assistants pass it over? The test, and with it the usage that would have grown on your platform instead of a rival’s. Buyers are actively shopping. Deepgram found 84% of respondents plan to increase voice budgets within a year and only 21% are very satisfied with their current voice technology. Gartner found 44% of service leaders exploring a voicebot but only 5% with one deployed, which means most contact center vendor choices in this category have not been made yet. A vendor missing from the answers at this stage misses the evaluation itself. Usage pricing raises the stakes. A developer who builds on a rival’s API and ships pays that rival for every character or minute afterward, and switching means re-testing voices, latency and integrations. That switching cost is our inference from how these products are priced; no study has measured it for voice. For a large brand, the risk runs the other way: a16z’s ranking and ElevenLabs’ growth show how concentrated attention can become, and our article on [whether AI assistants favor big brands](https://underneath.agency/resources/do-ai-assistants-favor-big-brands) explains why challengers need a clearly stated advantage to break in. ## What does GEO involve for a text to speech or voice agent company? It gives assistants clear, checkable facts about voices, latency, pricing and consent; it cannot secure a place in any answer. For a voice company, generative engine optimization (GEO) tends to span seven pieces of work: 1. **One clear description per buyer.** Say plainly whether you sell text to speech, voice cloning, dubbing, speech recognition or voice agents, and for which use cases, the same way on your site, docs, marketplace listings and profiles. 2. **Independent coverage and benchmarks.** Developer write-ups, latency and quality comparisons by third parties, customer case studies, analyst and trade coverage in customer service publications. These are the sources AI search leans on. For how a voice brand earns that coverage, read [how brands build authority for AI search](https://underneath.agency/resources/how-brands-build-authority-for-ai-search). 3. **Comparison pages with real numbers.** Latency, languages, voice quality samples and per-minute cost against named alternatives, with test conditions. Whether such pages get cited is covered in [our summary on comparison pages](https://underneath.agency/resources/do-comparison-pages-help-b2b-ai-citations). 4. **A pricing page an assistant can quote.** Explain credits in plain units (characters, minutes, calls), show what telephony and model costs add, and retire old price tables. 5. **Consent, safety and legal pages.** How you verify consent for cloning, what you block, how output is watermarked, and how you support FCC, ELVIS Act and EU labeling obligations, on public pages with dates. 6. **Readable developer documentation.** Public, current API docs and quick-start guides, since developers and their coding assistants read them directly. Our [agent-readable web study](https://underneath.agency/research/agent-readable-web-study) found only 3.2% of top websites return Markdown when an AI agent asks for it. 7. **Tracking voice questions in every major assistant.** Track a fixed set of use-case, technical, price and consent questions in ChatGPT, Gemini, Perplexity, Copilot and Google, and ask new accounts how they found you. ## What is still unknown about how voice buyers use AI assistants? How often they pick a voice vendor from an AI answer, and what usage follows: nobody has published it. The market figures here come from vendors (ElevenLabs, Deepgram) and investors (a16z) with an interest in the category’s growth, and the Gartner figures are forecasts and survey results, not measured outcomes. What we know about how AI search chooses software rests on general software questions and Product Hunt startups; none of it isolates voice tools. No study yet shows how assistants answer consent or legal questions about voice cloning, or whether published safety changes alter those answers. To judge whether voice visibility work is paying off in usage, read [our article on business results](https://underneath.agency/resources/does-ai-visibility-drive-business-results). ## How can a voice company see which AI answers feed its usage revenue? Check what assistants say about you for the use cases behind your biggest usage accounts. A practical first step is an audit of your use-case, technical, pricing and consent questions across the main assistants, matched against where your signups, API usage and contact center deals come from, so the gaps that cost the most usage revenue get fixed first. To have us map those answers against your API usage and contact center pipeline, [ask us for a voice visibility audit](https://underneath.agency/contact). Our [generative engine optimization service](https://underneath.agency/services/generative-engine-optimization) page covers what happens after that audit: clearer consent, latency and pricing pages, readable developer docs and ongoing measurement. ## Frequently asked questions ### Do AI assistants recommend the voice tool with the best audio quality? Not necessarily. Assistants do not say how they pick a voice tool, and research on software questions shows AI search leaning heavily on independent sources. A product with excellent audio but little independent coverage can be well known by name yet rarely suggested. ### Will negative coverage about voice cloning safety show up in AI answers? It can. Answers draw on independent sources such as consumer groups and the press. The practical response is to publish dated, specific information on what your safeguards are now, so newer and clearer sources exist alongside older criticism. ### Is usage-based pricing a problem for AI answers? It makes errors more likely. Credits, per-minute rates and telephony charges are hard to summarize, and assistants often quote prices imperfectly. A plain pricing page with worked examples reduces the room for error. ### How can a voice company tell whether customers come from AI search? Use three signals together: AI referral traffic, a “how did you hear about us?” field on the signup or API key form, and repeated checks of what assistants say for your main voice use cases. Referrals on their own miss much of it, since many AI-driven visits show up as homepage or direct traffic. ## Sources - ElevenLabs (2026), [ElevenLabs raises $500M Series D at $11B valuation](https://elevenlabs.io/blog/series-d) - ElevenLabs (2025), [Series C announcement](https://elevenlabs.io/blog/series-c) - ElevenLabs (2026), [Pricing](https://elevenlabs.io/pricing) and [Safety](https://elevenlabs.io/safety) - Andreessen Horowitz (2025), [AI Voice Agents: 2025 Update](https://a16z.com/ai-voice-agents-2025-update/) - Andreessen Horowitz (2026), [The Top 100 Gen AI Consumer Apps, 6th Edition](https://a16z.com/100-gen-ai-apps-6/) - Deepgram (2025), [State of Voice AI 2025](https://deepgram.com/learn/state-of-voice-ai-2025) - Gartner (2024), [Gartner Survey Reveals 85% of Customer Service Leaders Will Explore or Pilot Customer-Facing Conversational GenAI in 2025](https://www.gartner.com/en/newsroom/press-releases/2024-12-09-gartner-survey-reveals-85-percent-of-customer-service-leaders-will-explore-or-pilot-customer-facing-conversational-genai-in-2025) - Gartner (2022), [Gartner Predicts Conversational AI Will Reduce Contact Center Agent Labor Costs by $80 Billion in 2026](https://www.gartner.com/en/newsroom/press-releases/2022-08-31-gartner-predicts-conversational-ai-will-reduce-contac) - Gartner (2025), [Gartner Predicts Agentic AI Will Autonomously Resolve 80% of Common Customer Service Issues Without Human Intervention by 2029](https://www.gartner.com/en/newsroom/press-releases/2025-03-05-gartner-predicts-agentic-ai-will-autonomously-resolve-80-percent-of-common-customer-service-issues-without-human-intervention-by-20290) - Retell AI (2026), [Pricing](https://www.retellai.com/pricing) - Vapi (2026), [Pricing](https://vapi.ai/pricing) - Consumer Reports (2025), [Consumer Reports’ Assessment of AI Voice Cloning Products](https://www.consumerreports.org/media-room/press-releases/2025/03/consumer-reports-assessment-of-ai-voice-cloning-products/) - Federal Communications Commission (2024), [FCC Makes AI-Generated Voices in Robocalls Illegal](https://docs.fcc.gov/public/attachments/DOC-400393A1.pdf) - State of Tennessee (2024), [Gov. Lee Signs ELVIS Act Into Law](https://www.tn.gov/governor/news/2024/3/21/photos--gov--lee-signs-elvis-act-into-law.html) - EU Artificial Intelligence Act (2024), [Article 50: Transparency Obligations](https://artificialintelligenceact.eu/article/50/) - G2 (2026), [In the Answer Economy, Don’t Win the Click — Win the Answer](https://company.g2.com/news/g2-research-the-answer-economy) - Similarweb (2026), [AI Search Stats 2026: Market Share, Referral, and Citation Data](https://aisearch.similarweb.com/blog/gen-ai-stats/) - Chen and colleagues (2025), [Generative Engine Optimization: How to Dominate AI Search](https://arxiv.org/abs/2509.08919) - Sharma (2025), [The Discovery Gap: How Product Hunt Startups Vanish in LLM Organic Discovery Queries](https://arxiv.org/abs/2601.00912) - Underneath (2026), [software pricing accuracy](https://underneath.agency/research/ai-pricing-accuracy-study) and [agent-readable web](https://underneath.agency/research/agent-readable-web-study) --- This is the Markdown twin of https://underneath.agency/resources/ai-voice-software-revenue-from-ai-search. The HTML page is canonical. Publisher: Underneath, https://underneath.agency/agent. Site index: https://underneath.agency/llms.txt.