AI Answering Service Buyer's Checklist: What to Check Before You Choose One

Why "AI answering service" doesn't mean one thing

Four vendors can all market themselves as an AI answering service and still be selling four different products. Before comparing a single price or feature, it's worth separating what "AI" actually does in each one, because that changes what a fair price comparison even looks like.

XBert, Nextiva's AI receptionist product, is a standalone AI agent: it answers the call itself, books the appointment itself, and only brings in a person when the request needs one. Nextiva's own product page describes it as an "AI employee" that "greets callers with a professional, customizable voice, handles common questions using your knowledge base," and hands off to a team member only when needed, with full conversation context attached [nextiva.com/products/ai-receptionist, checked 2026-09-23]. Goodcall works the same way structurally: a self-configured AI agent tied to its own phone number, with a default fallback that transfers to the business line only when the agent can't handle a request [goodcall.com, checked 2026-09-23]. Smith.ai sells both models under one brand, an AI Receptionist that answers calls with AI, and a separate Virtual Receptionist product staffed by live agents, and the two are priced and structured differently enough that "Smith.ai's price" isn't a single number [smith.ai/pricing/ai-receptionist and smith.ai/pricing/receptionists, checked 2026-09-23].

Ruby is the one worth reading closely before assuming it belongs in this comparison at all. Ruby's own AI page states the company's position directly: "Ruby does not use AI to... Replace human connections. People prefer talking to people. Ruby's AI tools are designed to support and remove barriers from that tried-and-true form of connection rather than needlessly disrupting it." [ruby.com/ai, checked 2026-09-23] Every Ruby plan is answered by a trained human receptionist; the AI layer (Call Nav, which surfaces business information to the receptionist mid-call, call and voicemail transcripts, sentiment analysis, and robocall filtering) supports the agent [ruby.com/ai, checked 2026-09-23]. That's a legitimate answer to "how do I handle after-hours calls," and for some businesses, the right one. But it's a different product category from the other three. The checklist item that matters is whether you actually want a human on the line, or you're comparing Ruby on price against three AI agents it isn't structurally competing with.

Price-per-call, side by side

The unit each vendor bills by is the first thing to check, because it changes which business it fits. XBert and Smith.ai's AI Receptionist charge per interaction and per call, so a caller who phones back five times in a week generates five billable events on both. Goodcall bills per unique caller per month instead, one phone number that calls back repeatedly and actually talks to the agent counts once, which favors a business with a lot of repeat callers over one with mostly first-time contacts. Ruby bills by the minute, the only one of the four where a long, complicated call costs materially more than a short one, since Ruby is timing a live human conversation with human receptionists [ruby.com/plans-and-pricing, smith.ai/pricing/ai-receptionist, goodcall.com/pricing, nextiva.com/products/ai-receptionist, all checked 2026-09-23].

Two billing details need checking. First, Smith.ai's AI Receptionist pricing shown above is current as of this check and is meaningfully cheaper, on a per-call basis, than the per-minute Virtual Receptionist plans Smith.ai also sells under the same brand ($300/mo for 30 calls, roughly $10/call, on the human-staffed side) [smith.ai/pricing/receptionists, checked 2026-09-23]; if you're quoted a Smith.ai price, confirm which of the two products it's for. Second, Goodcall currently lists Starter at $79/month for 100 unique customers and $0.79 for each additional customer; confirm how multiple numbers or locations will be configured before assigning a separate subscription to each site [goodcall.com/pricing, checked October 11, 2026]. For a deeper look at the line items that inflate a quoted answering-service price after you sign, rounding rules, billing-cycle quirks, and per-transfer surcharges, see our answering service cost breakdown, which covers Ruby's and Smith.ai's billing fine print in more detail than fits here.

Setup time: minutes, days, or weeks

The AI-first products in this comparison advertise setup measured in minutes, and that claim holds up on the vendors' own sites, with one caveat: "minutes to configure" and "minutes to be fully trained on your business" aren't the same thing.

Nextiva describes XBert's setup as three steps, XBert "learns from your website, FAQs, and business details automatically," you pick a voice and set call-routing rules, then go live, and states "most businesses go live in minutes," with assisted setup from Nextiva's team included at no extra charge [nextiva.com/products/ai-receptionist, checked 2026-09-23]. Goodcall describes an equivalent process, connect a website or Google listing, customize behavior, get a number or forward an existing line, explicitly marketed as needing no API work or developer involvement [goodcall.com, checked 2026-09-23]. Test both claims directly: ask to see the agent's first few responses on your actual FAQs before deciding the automatic setup got your business right.

Smith.ai's AI Receptionist doesn't publish one canonical setup-time figure on its own site, but independent reviews and comparison sites converge on a self-serve configuration window of roughly 15 minutes to under an hour, with full go-live, once a number is forwarding correctly, typically landing within a day or two [aggregated from multiple third-party comparison sources, checked 2026-09-23]. Ruby and Smith.ai's Virtual Receptionist (human) product both run on a different clock, because a company is training people on your specific call-handling preferences, not configuring software: expect 24 to 48 hours to go live using standard templates, extending to three to five business days for a setup with multiple integrations or a detailed intake script, and some users report it taking a week or two to get the handling fully dialed in [ruby.com/get-started and ruby.com/faqs, checked 2026-09-23; aggregated third-party sources]. Neither pace is wrong, a live receptionist genuinely needs training time an AI configuration doesn't, but budget your actual go-live date around the product you picked, not the fastest number in the comparison.

Industries: named fit versus general-purpose

Ask a vendor directly whether they've configured an agent for a business like yours before, and check that against what they publish, because the answer changes how much manual tuning you should expect to do yourself.

Nextiva lists specific industries for XBert, Home Services (HVAC, plumbing, electrical, landscaping, pest control), Healthcare & Dental, Professional Services (law, accounting, financial advisors, consulting), Restaurants & Hospitality, Fitness & Wellness, and Retail & E-commerce, each with a short line about what the agent handles in that vertical [nextiva.com/products/ai-receptionist, checked 2026-09-23]. Smith.ai's site lists a similar spread by name: Legal, Home services, Software & IT services, Marketing agencies, General contractors, Finance, Medical & wellness, Hiring & staffing, and Insurance [smith.ai, checked 2026-09-23]. Ruby's "Who we serve" section names Legal, Home services, Business services (marketing, IT, insurance, design agencies), Healthcare, Financial services, and small businesses/freelancers/startups generally [ruby.com, checked 2026-09-23].

Goodcall is the exception: its own site doesn't publish a comparable named-industry list. Its use-case content instead points at local, appointment-driven service businesses, salons, schools handling parent communication, and general small-business call volume, plus a business-process-outsourcing angle for larger operators, but there's no single page laying out verticals the way the other three do [goodcall.com, checked 2026-09-23]. That's not necessarily a gap: Goodcall markets itself as configurable across industries, and its pricing and setup are the same regardless of industry. It does mean a Goodcall buyer should ask directly for examples close to their own business to verify fit, where the other three published vendor marketing already does some of that legwork.

Verify the healthcare claims: Nextiva markets XBert for Healthcare & Dental with language about maintaining "HIPAA compliance," and Goodcall separately advertises "HIPAA compliance software" as part of its security posture [nextiva.com/products/ai-receptionist and goodcall.com, checked 2026-09-23]. Neither page states, in the text reviewed here, that a signed Business Associate Agreement is included by default or how to request one. As our HIPAA compliant answering service checklist covers in more detail, a "HIPAA-compliant" label on a marketing page is not the same legal fact as a signed BAA, and that checklist's existing research already confirms Ruby offers a free opt-in BAA on request. If patient information will touch any of these tools, get the BAA question answered directly and in writing before routing a single call through it, regardless of which vendor's marketing copy uses the word.

What escalation actually costs

An AI agent's advertised per-call or per-interaction price is not the ceiling on what a bad month costs, because every one of these products eventually needs to hand a caller to a person, and that handoff sometimes carries its own charge.

Smith.ai's Virtual Receptionist (human-staffed) plans list a $1.50 per-call charge for appointment booking and a $0.50 charge for SMS/Slack transfer notifications as line items on top of the base plan [smith.ai/pricing/receptionists, checked 2026-09-23]; those are listed on the pricing page, which counts in the vendor's favor even though it adds to the real bill. Smith.ai's AI Receptionist pricing page doesn't list a comparable transfer or escalation fee of its own, so that disclosure may not carry over to the AI-first product this comparison is actually about. XBert's pricing page states a flat $99/mo covers "smart handoff to your team members" as one of six included features, with no separate transfer fee called out in the materials reviewed [nextiva.com/products/ai-receptionist, checked 2026-09-23]. Goodcall's default behavior transfers an unresolved call to the business line at no listed extra charge, since Goodcall bills by the monthly count of customers served [goodcall.com, checked 2026-09-23].

Ruby's live agents are the call by default, so "escalation cost" doesn't apply the same way, but Ruby's minute-based billing means a long, complicated call that a self-service AI agent might resolve in 90 seconds instead runs Ruby's per-minute meter for the full conversation. Ask each finalist to walk through, specifically, what happens on a call the AI (or the receptionist) can't fully resolve: does it transfer, take a message, or offer a callback, and is there a dollar figure attached to that specific outcome. Asking that question in a trial is the more reliable version of the same question.

What review data can and can't tell you

Independent review scores are useful for spotting a pattern, they're less useful as a precise ranking, and the sample sizes here vary enough to matter.

Smith.ai's G2 listing is split across multiple product profiles: Virtual Receptionist sits around 4.6-4.9/5 across roughly 73 reviews, and a separate AI Receptionist profile sits around 4.7/5, meaning a single "Smith.ai's G2 score" doesn't exist without specifying which product [aggregated from search-indexed G2 data, checked 2026-09-23]. Ruby shows a notable split between platforms: roughly 3.7-3.8/5 on G2 from a small sample of 11-12 reviews, against 4.6/5 on Trustpilot across a much larger sample of around 830 reviews [aggregated from search-indexed G2 and Trustpilot data, checked 2026-09-23]. A double-digit G2 sample can swing several tenths of a point from a handful of reviews either way, so treat the Trustpilot figure, with its larger sample, as the more stable read on Ruby specifically, while still checking G2's written reviews for the kind of detail a star rating alone won't show.

XBert doesn't have an independent review history of its own yet. It's a newer product. Nextiva displays company-wide Trustpilot, G2, and Gartner scores instead (4.5-4.8/5 depending on which badge on the page you read, with review counts in the thousands), and those reflect Nextiva's broader phone and CX business, not XBert specifically [nextiva.com/products/ai-receptionist, checked 2026-09-23].

For Goodcall, the review picture is unconfirmed. Search results returned one claim of a 4.8/5 G2 score across 1,400-plus reviews, and a separate claim that Goodcall's G2 profile has zero reviews and has sat inactive. A direct attempt to check G2 directly was blocked by the site's bot protection. The conflicting figures mean that Goodcall's G2 standing could not be independently confirmed here, and a prospective buyer should pull it up directly before treating any third-party summary as settled. One specific, attributed Trustpilot review for Goodcall does describe a billing dispute: continued charges after an October 2025 cancellation that the reviewer says went unresolved for months. Treat it as one account, not a verdict on the company's billing practices generally, the same way this site's existing cost breakdown treats a single outlier complaint about another vendor.

A short test before you commit to any of them

Whichever of the four fits your checklist on paper, run the same five-minute test before signing a contract: call the number yourself, ask a question that isn't in the vendor's demo script, and see what happens when the agent (human or AI) doesn't have a scripted answer ready. For the AI-first products, XBert and Goodcall, that tells you how gracefully the system escalates an unfamiliar question. For Smith.ai's AI Receptionist, it's a chance to see the $2-3/call overage rate in context against how often a real caller's question falls outside the plan. For Ruby, it's the more traditional version of the same test, listening for whether the receptionist sounds like they actually know your business or are working from a generic script.

None of that replaces the checklist items above, pricing unit, setup timeline, named industry fit, escalation cost, and review sample size, but a live test call catches the gap between a vendor's demo, which is built to go well, and the version of the product your actual callers will experience on a normal Tuesday.

Frequently asked questions