A year or two ago, the question about AI phone agents was whether they could hold a conversation at all. By late 2026 that question is mostly settled, and buyers ask about other things: how fast it answers, what happens when the caller switches to Spanish, what it can finish without a person, and what you're required to tell callers.
These are the five changes we'd point to. Each one is something you can check yourself on a single test call, and where a claim depends on research or regulation, the source is linked.
1. Replies now come in under a second
People are fast at taking turns. A study of conversations in ten languages, from English and Japanese to indigenous languages of Mexico and Namibia, found that the most common gap between a question and its answer was somewhere between 0 and 200 milliseconds (Stivers et al., PNAS, 2009). That's the rhythm callers expect without thinking about it.
A voice agent that pauses two or three seconds after every question breaks that rhythm. It sounds like a dropped line, so callers repeat themselves or talk over it, and the call falls apart. The models underneath have closed most of the gap: OpenAI's GPT-4o system card, for example, reports responses to audio in as little as 232 milliseconds, with an average of 320 (OpenAI, 2024). A phone agent has more to do than the model alone, since it has to hear the caller finish, look up your information and speak, but a well-built one now answers inside a second. ModPhone's does, on voice.
How to check: ask a vendor's demo line a question and count the silence before it starts to answer. Then interrupt it halfway through a long answer and see whether it stops and listens.
2. The caller picks the language, mid-sentence if they like
According to the Census Bureau, more than one in five people age 5 and older in the U.S. (22%) spoke a language other than English at home in 2017 to 2021 (U.S. Census Bureau, June 2025). The traditional answer was a "press 2 for Spanish" menu or a separate line, which assumes the caller picks one language at the start and sticks with it.
Real calls don't go that way. A parent starts in English, gets stuck on a detail about enrollment and switches to Spanish. A grandmother hands the phone to her grandson. A good agent detects the language from what the caller says and follows the switch. ModPhone's agent covers more than 30 languages and switches mid-call without any setting. Greenville Humane Society, which handles more than 1,500 calls a month with ModPhone, answers in English and Spanish around the clock, and a large K-12 district's after-hours parent line on ModPhone is bilingual too.
How to check: start a question in English and finish it in another language.
3. The agent finishes the task instead of taking a message
The traditional model, at many answering services and in early phone bots, was to take a message and hand the work back to you. You still had to call the person back to book the appointment or answer the question they called about.
The shift is toward agents that do the work through the tools you already use. On a single call, ModPhone's agent can check your Google or Outlook calendar and book the appointment, then log the call as a row in Google Sheets or on the matching HubSpot contact. It can also send the summary and transcript to any other system through a webhook, which is how it connects to Zapier, Make and n8n. The caller hangs up with the thing done, and you get a summary instead of a to-do.
The question to ask here is less "does it integrate with X" and more "what exactly does it write, and what does it never touch?" A good vendor can tell you precisely. ModPhone's calendar connection, for example, adds its own appointment events and never edits or deletes events you created.
4. Links get texted during the call
Spelling a web address over the phone rarely works. By "h, t, t, p, s, colon, slash, slash" the caller has already lost it. An agent that can text sends the link instead, while the caller is still on the line: the enrollment page, the client portal, the payment page, directions to the office.
ModPhone's agent offers to text things that are easier to read than hear, asks whether to send it to the number the caller is using or a different one (landlines can't receive texts), and confirms once it's sent.
The wireless industry's own guidelines suggest a rule of thumb. CTIA's Messaging Principles and Best Practices describe informational messages as ones a consumer agreed to receive for a specific purpose when they gave their number, and note that adding a call to action such as a coupon code can move a message into the promotional category, where the consumer should agree in writing first. Keep in-call texts to exactly what the caller asked for.
5. Rules on AI voices and disclosure are taking shape
This is the change most likely to affect how you set up an agent, so it's worth being precise about what the rules say.
In February 2024 the FCC ruled that AI-generated voices count as "artificial" voices under the Telephone Consumer Protection Act (FCC 24-17). The ruling is about calls a business places: an outbound call using an AI voice generally needs the called party's prior express consent, the same as any prerecorded-voice robocall. In August 2024 the FCC also proposed requiring callers to disclose AI-generated calls (proposed rule). As of this writing, no final rule on that proposal has been published in the Federal Register.
States have moved faster, and their rules are not limited to outbound calls. Utah's 2025 amendments require a business using generative AI with a consumer to disclose it when the person clearly asks whether they're talking to a human or an AI, and require state-licensed professions to disclose it at the start of high-risk interactions, such as collecting health or financial data or giving financial, legal or medical advice (Utah S.B. 226). Maine's law, signed in June 2025, prohibits using a chatbot, including one that speaks, in a way that may mislead a reasonable consumer into thinking they're dealing with a person, unless the consumer is clearly told otherwise (Maine Public Law 2025, ch. 294). And in the EU, the AI Act's transparency obligations, which include informing people when they're interacting with an AI system, apply from 2 August 2026 (European Commission).
The simple way to stay on the right side of all of these is to say it up front. ModPhone's default greeting already does: "This is [Agent Name], your AI assistant." If you rewrite the greeting, keep that part. This isn't legal advice; if you make outbound calls or work in a regulated profession, check with your counsel.
What didn't change
Some calls still need a person, and that hasn't moved. A caller who's upset, a question that needs judgment or access to a private record, a genuine emergency: the agent's job there is to get the caller to the right human quickly, with the details already written down.
And an agent is still only as good as what it's been told. The best voice and the fastest reply don't help if the knowledge has last year's hours in it. Most of the work of a good setup is keeping facts current.
Common questions
Do I have to tell callers they're talking to an AI?
It depends on where you and your callers are and what kind of business you run. Utah requires disclosure when a caller clearly asks, and up front for some regulated professions; Maine prohibits AI that could mislead a reasonable consumer into thinking it's human unless they're told. The safest practice everywhere is to say so in the greeting, which ModPhone's default greeting does.
Does the FCC's AI robocall ruling apply to an AI receptionist answering my calls?
The February 2024 ruling applies the TCPA's consent rules to calls placed using an AI-generated voice, which means outbound calling. An AI receptionist answering calls people make to you is a different situation, though state disclosure laws can still apply. If you plan to place calls with an AI voice, you need the recipient's prior express consent.
How fast should an AI receptionist respond?
Within about a second of the caller finishing. In ordinary conversation people usually reply within a fraction of a second, so longer pauses sound like a problem on the line. ModPhone's agent responds in under a second on voice.
How many languages should an AI receptionist speak?
Enough to cover your callers, and it should switch on its own. ModPhone's agent handles more than 30 languages and follows the caller if they change language mid-call, with no menu or setting.
Try it on one call
To see where a product stands on all five, call it once. Try the ModPhone demo line at +1 (234) 231-6622, or paste your own website into the live demo.