In five lines
- Scores come from five criteria with written rules. Anyone can apply the rules and check our work.
- Every price, limit and percentage is stored once, with its source and the date we read it. Every page in every language is generated from that record.
- Today the rubric rates public documentation: pricing pages, help centers, trust pages. It does not rate how good a bot's answers are.
- We have designed a reference-bot build test. It has not been run. Until it has, nothing on this site says a product was tested.
- No affiliate links, no paid placement. Rankings follow the rubric scores, and no product can buy a better position.
What kind of review this is
Most chatbot roundups are assembled from vendor landing pages and from each other, and then described as testing. We would rather be exact about what stands behind a page, even when the answer is less impressive.
What we do today is desk research. We read each vendor's pricing page, help center, API documentation, terms, trust page and sub-processor list. We read case studies that name the customer. We read platform rules from Meta, because they decide what a WhatsApp or Instagram bot may do. We write down what each source says, link to it, and score products on what a buyer can verify before paying.
What we have not done is build bots. We have not yet built our reference bot on these platforms, so we publish no build times, no answer-quality results and no containment rates of our own. The protocol for that work is public on the reference-bot test page, along with its status.
Where facts come from
- The vendor's own pricing page, documentation and legal terms for that vendor's prices, metered units, limits, channels, hosting and certifications.
- The platform owner's policies (Meta for WhatsApp, Messenger and Instagram) for what bots may do on those channels and what the platform charges.
- Case studies that name the customer for results in production, read knowing that vendors choose which customers to publish.
- Independent reporting, regulators and the text of the law for everything a vendor has an interest in describing generously.
One database, so pages cannot disagree
Review sites often show one price in a ranking and another in a comparison, because each article was typed separately. Here, prose never contains a number. It points to a record in the product database, and the record holds the value, the source and the date. Reviews, pricing pages, comparisons, the chatbot finder and all four language editions are generated from it. A price change is made once and appears everywhere.
Prices are shown in the currency the vendor publishes. Most vendors here publish in US dollars and some in euros. We never convert, because a converted price is not one you can actually pay.
How we score
Rubric v1.0 has five criteria. On each one a product gets a level from 1 to 5, using the written rules below; a product gets the highest level whose rule it fully meets. Levels are multiplied by the weights and summed to a score out of 10. If a score does not follow from the rules, tell us and we will fix it.
Pricing clarity — 25%
Can you work out next month's bill from the vendor's own pages?
| Level | What it means |
|---|---|
| 5/5 | The full price list is public, the metered unit is defined in writing, and the vendor states what happens and what it costs when you pass the allowance. |
| 4/5 | Plan prices and allowances are public and the metered unit is defined in writing, but one material part of the bill (AI usage, overage or a required add-on) is priced only on request or passed through at a variable cost. |
| 3/5 | Plan prices are public, but the metered unit is not defined or the price beyond the allowance is not published. |
| 2/5 | Quote only, but the vendor documents what it meters. |
| 1/5 | No prices and no public description of how the vendor charges. |
Access: free plan and time to a first bot — 15%
Can you build and publish a bot before you pay or talk to anyone?
| Level | What it means |
|---|---|
| 5/5 | A permanent free plan that can run a live bot, with its limits published. |
| 4/5 | A permanent free plan, but it cannot run a live bot in practice (no publishing, no AI, or a cap too low for a small site). |
| 3/5 | A time-limited free trial with self-serve sign-up; no permanent free plan. |
| 2/5 | A demo or a sandbox, but a bot cannot be published without a contract. |
| 1/5 | Sales-led only: no trial and no self-serve sign-up. |
Builder depth and control — 20%
How far can you take a bot before the tool stops you?
| Level | What it means |
|---|---|
| 5/5 | Everything in level 4, plus a choice of AI model or your own key, and a documented way to export your bot. |
| 4/5 | Everything in level 3, plus custom code or an SDK, a test emulator and versioning or environments. |
| 3/5 | Visual flows plus AI answers from your own content, with API calls or webhooks. |
| 2/5 | A visual flow builder or an AI agent, but not both, and no way to call your own systems. |
| 1/5 | Fixed templates or FAQ answers only. |
Reach and human handoff — 20%
Does it work where your customers are, and can a person take over?
| Level | What it means |
|---|---|
| 5/5 | Everything in level 4, plus voice or SMS, and AI answers documented in German and Spanish. |
| 4/5 | Website, WhatsApp and at least one Meta channel, with a built-in inbox or documented help-desk handoff. |
| 3/5 | Website plus at least one messaging channel, with documented handoff. |
| 2/5 | One or two channels, with handoff only through a third-party tool. |
| 1/5 | One channel and no documented handoff to a person. |
Data, compliance and evidence — 20%
Can you pass a security review and check the vendor's claims?
| Level | What it means |
|---|---|
| 5/5 | Everything in level 4, plus documented EU data hosting or self-hosting. |
| 4/5 | A DPA, SOC 2 or ISO 27001, a public sub-processor list and named-customer case studies with figures. |
| 3/5 | A DPA and a stated hosting region, or SOC 2 / ISO 27001, but not both. |
| 2/5 | A GDPR statement or DPA, but no certifications and no stated hosting region. |
| 1/5 | No security or data-processing documentation found. |
The ranking
| Product | Pricing clarity | Access: free plan and time to a first bot | Builder depth and control | Reach and human handoff | Data, compliance and evidence | Score |
|---|---|---|---|---|---|---|
| Intercom Fin | 5/5 | 3/5 | 4/5 | 5/5 | 3/5 | 8.2 |
| Botpress | 4/5 | 4/5 | 4/5 | 4/5 | 4/5 | 8.0 |
| HubSpot Chatbot Builder | 5/5 | 4/5 | 3/5 | 4/5 | 3/5 | 7.7 |
| Landbot | 4/5 | 5/5 | 3/5 | 4/5 | 3/5 | 7.5 |
| Tidio Lyro | 4/5 | 5/5 | 3/5 | 4/5 | 3/5 | 7.5 |
| Manychat | 5/5 | 4/5 | 3/5 | 2/5 | 3/5 | 6.9 |
| Typebot | 4/5 | 5/5 | 3/5 | 2/5 | 3/5 | 6.7 |
| Voiceflow | 2/5 | 4/5 | 5/5 | 3/5 | 3/5 | 6.6 |
| Chatfuel | 3/5 | 4/5 | 2/5 | 4/5 | 3/5 | 6.3 |
What the score leaves out
The score says how clear the pricing is, how easy it is to get started, how far the builder goes, where the bot can run and how well the vendor documents its handling of data. It does not say whether the bot will answer your customers correctly. That depends on your content and your questions, and only a build on your own material will show it. Our guide to writing a bot knowledge base is the best place to start.
The rubric also puts very different tools on one scale: a marketing bot for Instagram, a developer platform and a support agent are not substitutes. That is why each best list re-reads the same data for one job, and why a lower overall score can still be the right pick for yours.
What changes when the build test is run
Rubric v2 will add measured criteria from the reference-bot test: answer quality, build time and handoff behavior. The weights will be published before the first run, not after, so results cannot shape them. Products that cannot be tested without a sales contract will keep a documentation-only score and be labeled that way.
Keeping pages current
Each page shows when it was published, when it was updated and when its prices were last checked against their sources. We re-check the database monthly and whenever a vendor announces a pricing change. Anything that could change a buying decision is logged in the changelog.
Independence, funding and AI assistance
Rankings are ordered by rubric score. Pages are researched and drafted with AI assistance from the sources listed on them. A named editor appears in the byline only when a real person has checked the page; we do not invent authors.
Corrections
Use the "Report an error" link on any page. Confirmed errors are fixed in the database, which corrects every page at once. Vendors are welcome to send corrections with a link to a public source; private assurances do not change a score.
Researched and drafted with AI assistance from the sources listed on this page. We have not built or tested bots on these platforms yet. Method: How we review