chatbots
By Saurav | Founder of saavos | Building in public toward $10k MRR
[!TLDR] Six platforms cover roughly 95% of the AI chatbot market in 2026: Chatbase, Intercom Fin, Wonderchat, Botsonic, Tidio, and saavos. The right choice almost never comes down to feature count — it comes down to setup speed, the underlying model, and whether the widget renders citations cleanly. Below is the head-to-head, the pricing reality check, and the four features nobody markets but everyone regrets ignoring.
Grade on the questions a real buyer asks before they pay — setup time, source support, underlying model, citation quality, starting price, billing transparency, and embed customization. The criteria above were chosen to reflect buyer trade-offs, not to favor any one product. Where competitors genuinely win, we say so. saavos wins on setup speed and model quality; Chatbase wins on source breadth and templates; Intercom Fin wins on escalation UX.
We graded each platform on the questions a real buyer asks before they pay:
Honesty disclaimer: saavos makes one of the platforms in this comparison. The criteria above were chosen to favor real buyer questions, not our product. Where competitors are genuinely better, we say so.
Six cover roughly 95% of the SMB market: Chatbase ($32/mo annual, most established), Intercom Fin ($99/mo + $0.99/resolution), Wonderchat ($49.99/mo, 80+ languages), Botsonic ($19/mo, bundled with Writesonic), Tidio (free to $29/mo, live chat included), and saavos ($19/mo, fastest setup, Claude Sonnet 4.6). Every other player is a variation on one of these. Pick based on your actual use case, not feature count.
| Platform | Setup time | Sources | Model | Citations | Starts at | Notes |
|---|---|---|---|---|---|---|
| Chatbase | <5 min | URL, PDF, Q&A, Notion | GPT-4 family / Claude | Inline + sources panel | $32/mo (annual) | The default choice. Mature, established, biggest template library. |
| Intercom Fin | 30+ min | Help center, Confluence, Zendesk, etc. | OpenAI | Inline | $99/mo per resolution | Best if you already pay for Intercom. Per-resolution billing scales weird. |
| Wonderchat | 5–10 min | URL, PDF, sitemap | GPT-4o, Claude | Inline | $49.99/mo | Strong multilingual marketing — 80+ languages out of the box. |
| Botsonic | ~10 min | URL, PDF, FAQ | GPT-4o | Inline | $19/mo | Part of the Writesonic suite; useful if you also need their content tools. |
| Tidio | ~10 min | URL, FAQ | Custom Lyro AI | Limited | Free → $29/mo | Live chat + bot in one. Lyro is purpose-built but smaller than frontier models. |
| saavos | 5 min | URL, PDF, text | Anthropic Claude Sonnet 4.6 | Inline + sources footer | $19/mo | Fastest setup. Anthropic-grade responses. Transparent flat pricing via Dodo as Merchant of Record. |
A useful third-party perspective is Chatimize's regularly-updated roundup of chatbot platforms — they grade for slightly different criteria (lead qualification, integration depth) and reach different rankings. Worth reading alongside this if you're earlier in the buyer journey.
Pick Chatbase if you want the safest, most documented option with the widest template library. Pick Intercom Fin if you're already paying for Intercom and need native ticket escalation. Pick Wonderchat if multilingual is a hard requirement. Pick Tidio if you need live chat alongside AI in one tool. Pick Botsonic if you're already on Writesonic. Pick saavos if you want the fastest setup with Claude-grade responses and a predictable flat bill.
You want the most established, most-documented option with the deepest template library. Chatbase has been around longest, has integrations with most of the SaaS tools you already use, and the dashboard is mature. It's the safest "no one got fired for buying" choice. The caveat is that it's also the most generic — your bot will look and behave a lot like everyone else's.
You already use Intercom for support. Fin slots right in: it inherits your help center articles, escalates to a human agent natively, and lives inside the Messenger your team is already trained on. If you're not already an Intercom customer, the per-resolution pricing is hard to predict and the standalone pricing is $99/month per resolution — one of the priciest in the category.
Multilingual support is non-negotiable — Wonderchat advertises 80+ languages and the dashboard UX itself is more international than most competitors. It's also a solid generalist; if you're outside the US-centric SaaS bubble, it's worth a closer look than its smaller market share might suggest.
You're already paying for the Writesonic suite for AI content. Botsonic plugs into the same dashboard, the same billing, and shares some of the underlying infrastructure. Standalone, it's a good Chatbase alternative; bundled, it's compelling.
You need live chat + bot in the same product. Tidio gives you human handoff, agent UI, and visitor analytics in one panel. The Lyro AI is purpose-built for support, which has trade-offs — it's tuned tightly but underperforms frontier models on edge cases.
You want the fastest path from URL to live bot, with Anthropic Claude Sonnet 4.6 quality and a clean embed widget. Pricing is flat ($0, $19, $49, $199 per month) with no per-message overage on the lower tiers, and Dodo Payments handles tax + invoices as Merchant of Record. The honest counterpoint: if you need 30+ pre-built templates or want a deep agent-handoff UI on day one, Chatbase or Tidio cover those better right now.
No — the same sticker price buys wildly different products. At $19/month, two platforms compete directly: Botsonic Starter gives you 5,000 messages on GPT-4o, while saavos Starter gives you 1,000 messages on Claude Sonnet 4.6 — half the volume but a meaningfully stronger model on multi-turn queries. Chatbase raised its entry paid tier to $32/month (annual billing) in early 2026, so it no longer sits in this $19 comparison. Test with your real visitor question mix, not a marketing demo prompt.
Sticker prices on chatbot platforms are easy to compare. What's harder — and what actually matters — is what those prices buy you in practice. The two $19/month plans are not the same product, and Chatbase at $32/month annual is a different buy decision entirely:
| Platform | Entry tier includes | Underlying model | Bots | Messages/mo | Price/mo |
|---|---|---|---|---|---|
| Chatbase Hobby | 2 chatbots, 4M training characters | GPT-4o (model selection avail.) | 2 | ~500 credits | $32 annual |
| Botsonic Starter | 5,000 messages | GPT-4o | 1 | 5,000 | $19 |
| saavos Starter | 2 bots, 1,000 messages | Anthropic Claude Sonnet 4.6 | 2 | 1,000 | $19 |
saavos's lower message ceiling reflects the model: Claude Sonnet 4.6 costs roughly 3-5× more per token than GPT-4o-mini, and Sonnet's response quality is meaningfully higher on multi-turn, ambiguous queries. Pick based on what your visitors actually ask. If you're getting 5,000 short FAQ-style questions a month and model quality isn't the constraint, Botsonic covers it at $19. If you're getting 1,000 complex support or sales questions, Sonnet's accuracy advantage compounds and saavos's ceiling is sufficient.
A good heuristic: estimate your monthly conversation volume, multiply by 6 (the average user-bot turn count for a real conversation), and that's the message count you actually need. Most teams overestimate.
Four: citation quality (inline superscripts that link to exact source paragraphs, not generic URLs), fallback message UX (what the bot does when it can't answer — hallucinate or route gracefully), embed widget customization (position, color, starter questions, bot name, mobile behavior), and operator dashboard depth (conversation review, source health, ability to mark wrong answers). Vendors lead with integrations; buyers regret skipping these four.
Citations are the single biggest difference between a chatbot people trust and one they dismiss. Look for inline superscripts that link to the exact source paragraph — not "here are 5 pages we read." Chatbase, Wonderchat, and saavos all do inline citations well. Watch out for platforms that show generic source URLs without anchoring them to specific claims; those citations are decorative, not load-bearing.
When retrieval finds nothing relevant, the bot should fail visibly and route the visitor somewhere useful. The default behavior on most platforms is to let the model "wing it" — which is how you get hallucinations. Configurable per-bot fallbacks ("Email support@yourbusiness.com — we reply within an hour") turn the failure mode from embarrassing into branded. Test this before launch by asking the bot something it absolutely shouldn't know, and seeing what it does.
The widget is what your visitors actually touch. Check that you can configure: launcher position (bottom-right vs bottom-left), launcher color, greeting message, 3–5 suggested starter questions, the bot's name in the chat header, and mobile behavior (full-screen vs panel). If a platform locks you into "blue launcher in the bottom right with our logo," it's not really embeddable — it's a hosted page in a fancy iframe.
After launch, you'll spend more time in the owner dashboard than the visitor will spend in the chat. Look for: conversation review (every chat, searchable, with the bot's confidence per turn), source health (which sources are being cited, which are dead weight), and a way to mark answers wrong so the bot retrains. Platforms that ship the chat well but skimp on the operator UX age badly.
Three buyer scenarios where the platform choice is genuinely load-bearing: a new launch where citation quality and fallback UX matter more than message volume; a legacy-to-AI migration where import fidelity and conversation review matter; and a scale-up where non-English traffic requires per-language retrieval, not prompt-based detection. In all three, the cost of retraining on a second platform after the first one fails exceeds the subscription by a factor of 10.
Picking the wrong chatbot platform in 2026 costs more than the subscription. The marginal cost is the time you spend retraining a second time when the first platform's retrieval underperforms, the trust you spend with visitors when the bot hallucinates in front of them, and the conversion you leave on the table when the embed widget reads "generic SaaS launcher" instead of "your brand." For a pre-revenue or early-revenue site, those costs compound fast. Three buyer scenarios where the platform choice is genuinely load-bearing:
The first is a new launch with no support history yet. You don't know what visitors will ask, your FAQ is half-finished, and your team is one or two people. The chatbot's job here is to make a polished first impression while you figure out what real support questions look like. That means citation quality and fallback UX matter more than message volume — you'd rather have 500 well-grounded interactions than 5,000 generic ones.
The second is migrating from a legacy support setup to AI-first. You already run a knowledge base, you already know your top 50 questions, and you're swapping out a $300/month live-chat seat for a $19–49/month bot. The platform choice rides on import quality (can it ingest your existing help-center articles cleanly?) and operator dashboard depth (can you review every conversation to verify the bot's not regressing on questions you used to answer well?).
The third is scaling beyond a single language or region. Multilingual chatbots in 2026 are uneven across platforms — some run a single model with prompt-based language detection, others train per-language indexes. If your traffic is non-English-majority, this is the single biggest filter and most platforms quietly underperform here. Test with three real visitor questions in your target language before paying.
Six mistakes, in order of frequency: picking by feature count instead of feature quality, ignoring the underlying model (GPT-3.5 vs Sonnet is not cosmetic), skipping mobile widget testing, not configuring a fallback message, treating the bot as set-and-forget after launch, and buying the second-from-top tier "just in case." The last one is the most expensive per incident — most teams overestimate their conversation volume by 3–10x.
After watching SMBs configure chatbot platforms across the major options, the same six mistakes show up over and over. Each is preventable but easy to miss until the bot is already live.
Picking by feature count, not by feature quality. A 50-feature comparison table on a vendor's marketing page tells you nothing about whether their citation handling works. The platforms with the most features on paper are frequently the platforms with the buggiest implementation of the three that matter (citations, fallbacks, retrieval threshold). Score the evaluation rubric in our checklist on three real questions per platform instead.
Ignoring the underlying model. A platform that runs GPT-3.5 on its entry tier and a platform that runs Anthropic Claude Sonnet 4.6 are not selling the same product. Test both with a multi-turn ambiguous question (the kind a real visitor would ask) and watch how each handles "I don't know" — that's where the model difference shows up, not in the marketing demo.
Skipping the embed widget on mobile. Most chatbot evaluations happen on desktop. Most visitor conversations happen on mobile. Test the launcher and the open chat at iPhone-12 width on every platform before paying — some widgets squeeze into a corner that's unusable on small screens, others go full-bleed gracefully. This is the most common post-purchase regret we see.
No fallback message. "I don't know" with no follow-up is a dead-end conversation. A configured fallback ("I'm not sure about that — email support@yourbusiness.com and we'll respond within the hour") converts a lost visitor into a captured lead. Most platforms support custom fallbacks; almost no one configures them. Set this before launch — it's the highest-leverage 30 seconds you'll spend.
Treating the bot as set-and-forget. A chatbot typically needs roughly 30 minutes of operator attention per week for the first month — reviewing conversations, marking answers wrong, tightening source scoping. Platforms that don't expose conversation logs make this impossible. The teams that hit 30–60% deflection rates run this loop; the teams that hit 10–20% don't.
Paying for tiers you don't need. The biggest cost mistake in chatbot platform selection is buying the second-from-top tier "just in case." Volume requirements are predictable: count last month's contact-form submissions, multiply by 4 (rough visitor-to-question ratio), and that's your monthly bot question volume. Most teams overestimate by 3–10x. Buy the tier that covers your real volume with 30% headroom — not the tier that would cover the volume you'd have if every marketing assumption pans out.
Seven companion posts cover what comes after you narrow your platform shortlist: a 12-point evaluation checklist, an AI vs live chat decision guide, step-by-step setup for common CMSes, an SMB-specific buyer guide with ROI math, a hallucination prevention guide, an honest list of what saavos doesn't do, and a 12-question vendor interview framework. Links below.
If this comparison helped you narrow the platform shortlist, these companion posts cover the next steps:
Use the free tier. Every platform in this comparison offers one — 50 conversations per month on saavos (no card), 1,000 AI messages free on Botpress, 50 Lyro conversations one-time on Tidio. Paste your FAQ or pricing URL, ask it 10 real visitor questions, and see what happens. The free tier test takes 20 minutes and eliminates 80% of the wrong picks. Start free on saavos — no credit card, upgrade in one click if it fits.
saavos is the fastest of these to set up — paste your URL, get a Claude-powered bot live in 5 minutes. Free plan available, no credit card. Start free or see our pricing to compare tiers side-by-side.
Get the next post in your inbox
Honest writing on building, embedding, and shipping AI chatbots. No spam. Unsubscribe anytime.
Tidio has the most generous free tier (50 conversations per month). For paid plans starting around $19, Chatbase, Botsonic, and saavos are roughly tied — but pay attention to message limits: $19 on saavos includes 1,000 Anthropic Claude messages, $19 on Chatbase includes 2,000 messages on a smaller model.
Tidio for live chat + bot combined. saavos for a no-credit-card forever-free tier with 1 bot, 50 messages per month, and the same Anthropic Claude Haiku model paying customers get on the lower tiers. Chatbase free is generous but expires after 30 days.
Yes — your sources (URLs, PDFs, text) live with you, not the platform. The lift is re-uploading the sources and replacing the embed script tag. Conversation history rarely transfers; export it via the platform dashboard before you cancel.
Accuracy depends more on retrieval quality than the underlying model. Industry consensus: platforms running Anthropic Claude Sonnet 4.5+ or GPT-4o on top of clean retrieval (proper chunking, citations, fallback tuning) all answer well. The differences show up in edge cases: ambiguous queries, multilingual content, and long conversations.
All six in this comparison work without code if you use their dashboard plus their official Webflow / Shopify / WordPress integrations. The "embed script" snippet still needs to be pasted into your site theme, but most platforms ship one-click installers for the major no-code builders.
Wonderchat and Intercom Fin have the most explicit multilingual marketing — Wonderchat advertises 80+ languages out of the box. In practice, any platform on a modern frontier model (Claude, GPT-4 family) handles major European and Asian languages well; the difference is whether the dashboard UX itself is translated.
Builds tools for solopreneurs and small SaaS teams who don't have an afternoon to spare.
Paste your URL. Train your bot. Drop one script tag. No credit card.