AI on your website: ten platforms compared.

We did this competitor analysis because we wanted to know where we stood, not to make ourselves look good. The scope is deliberately narrow — AI on your website, which is what copadre does — and one of the measures is one copadre doesn’t fully earn. We’ve left it in, and we say below what the table doesn’t measure.

The platforms

We picked on standing in the market, not on how they’d score: four enterprise leaders, four self-serve leaders, and one platform in between. Listed alphabetically, with copadre last.

Platform	Why it’s in
Ada	The longest-standing enterprise support-automation platform
Botpress	The best-known open-heritage builder; unlimited bots
Chatbase	The volume leader in self-serve website assistants
Intercom Fin	The market-leading outcome-priced agent; over 12,000 customers
Salesforce Agentforce	The most prominent “build any agent” platform
Sierra	The category leader by valuation and enterprise footprint — voice, chat, email, WhatsApp
Tidio (Lyro)	The largest small-business live-chat and AI agent install base
Voiceflow	The most complete builder platform; JPMorgan, Amazon
Zendesk AI agents	The incumbent helpdesk’s agent, now bundled into every Zendesk plan
copadre	Us

Close calls we left out: Decagon (an enterprise leader by any measure, but its documentation is gated and we couldn’t verify enough of it to score it fairly — we’d rather leave a row out than guess), Crisp, Kommunicate and Zoho SalesIQ (strong small-business tools with a smaller footprint than Chatbase or Tidio), CustomGPT, Lorikeet and DevRev. Swapping any of them in doesn’t change the shape of the table.

The criteria

We chose seven things that matter when you’re choosing, each of which you can check for yourself on the vendor’s own website, and each of which separates the field rather than giving everyone a tick.

Your own AI key. Full tick: you run the assistant on your own model-provider key and pay the provider directly, with no markup on AI usage. Half: only one provider’s key is accepted, or only on an enterprise tier.

Several assistants included. Full tick: the first paid plan includes more than one assistant without a per-assistant surcharge. Half: two to twenty, or bundled but priced per seat.

Works with your AI tools. Full tick: the vendor publishes an MCP server so you can manage the assistant from Claude, ChatGPT, Cursor or similar. Half: something partial — announced but not shipped, a local developer tool, a single client, or an action route rather than management.

Yours on your website. Full tick: a copy-paste snippet to install, no vendor badge on the deployed widget on any plan at no extra cost, and styling by your own CSS rather than a fixed set of theme options. Half: the snippet is there but a “Powered by” label costs an add-on or a higher tier to remove, or styling is limited to colours and themes, or deployment is an implementation project. We check this one carefully because it’s easy to miss until your assistant is live: most widgets in this market carry the vendor’s badge by default, which means your website is quietly advertising their AI service.

Choice of AI model. This one has two halves, and we score them separately. Can you choose which model each assistant runs on at all? Most platforms here don’t let you — Sierra, Ada, Zendesk and Intercom pick for you, and Tidio’s Lyro is fixed to one model. And if you can choose, is the choice across several vendors or within one? Full tick: you pick per assistant, across vendors (Chatbase, Voiceflow, Botpress). Half: you pick per assistant, within one vendor. copadre earns the half — you choose each assistant’s model from the current Claude range (Haiku, Sonnet, Opus), and only Claude.

Start without a sales call. Full tick: public prices and a real trial or free tier on the actual product. Half: a free tier that needs a parent platform, or a trial with the price list withdrawn.

Tells you what people are asking. Full tick: reporting that reads across conversations — recurring themes, your customers’ own phrasing, this period against last, what the assistant couldn’t answer, and written summaries — not just volumes and resolution rates. Half: outcome dashboards only, or topic analysis gated to a top tier or add-on.

Full ticks decide the order of the table. Where two platforms have the same number, the one with a mark in more columns goes first; if still level, alphabetical. Prices aren’t shown — you’ll find those on each vendor’s site, and you’ll decide whether they’re good value better than we can.

The scores, and where they came from

Every mark below was read from the vendor’s own pricing page, documentation or help centre on 27–28 August 2026. Platforms are in the same order as above — alphabetical, copadre last; the graphic orders its rows by score. Where a figure is only available from third parties we say so. Vendors change plans; if you spot something that’s moved, tell us and we’ll re-check.

Ada — ada.cx

Own key: no. Several assistants: no. AI tools: yes — a hosted MCP server for Claude, ChatGPT and Gemini covering configuration, behaviour, conversations and metrics. Your website: half — a genuine snippet install and no vendor badge that we could find, but styling is a fixed set of theme options with no custom CSS. Model: no. Sales call: no — demo only. Reporting: half — auto-generated Topics with example conversations; gaps found by manual review; no digests.

Sources: docs.ada.cx/docs/optimization/performance/topics/topic-analysis · ada.cx/platform.

Botpress — botpress.com

Own key: half — enterprise only. Several assistants: yes — unlimited bots on every paid plan. AI tools: half — a local developer-tool MCP server in the ADK; not a hosted connector. Your website: half — snippet install and the deepest styling of the ten (100+ CSS classes), but the free tier carries Botpress branding; a custom footer needs the first paid plan. Model: yes — multi-vendor. Sales call: yes — free tier, public pricing. Reporting: half — usage analytics; per-conversation AI summaries in beta; custom analytics on Team plans and above; no cross-conversation topics.

Sources: botpress.com/pricing · botpress.com/docs/studio/concepts/home · botpress.com/docs/changelog.

Chatbase — chatbase.co

Own key: no. Several assistants: no — one agent per plan, $25 a month each extra. AI tools: no. Your website: half — snippet install, but the widget says “Powered By Chatbase” unless you buy the removal add-on ($99 a month at time of writing — more than the Hobby plan itself); styling is theme options, no CSS. Model: yes — nine vendors. Sales call: yes — free tier, public pricing. Reporting: half — Topics, sentiment, knowledge-gap suggestions and a previous-period chart exist, but only on the Pro plan ($500 a month at time of writing); no digests.

Sources: chatbase.co/pricing · chatbase.co/docs/user-guides/chatbot/analytics.

Intercom Fin — fin.ai

Own key: no. Several assistants: no — one Fin. AI tools: half — MCP as an action route; Intercom’s connector is in Anthropic’s directory for helpdesk data. Your website: half — snippet install and polished theming, but “Powered by Intercom” stays on the Messenger until the Advanced plan ($85 a seat a month at time of writing), and there’s no custom CSS. Model: no — Intercom’s own models. Sales call: yes — 14-day trial, public pricing. Reporting: yes — Topics Explorer, Trends (each week against the prior twelve, with real customer questions), Recommendations, Fin Operator; on the Pro add-on ($99 a month at time of writing).

Sources: fin.ai/pricing · Intercom help centre articles 11390087, 11875255, 11390088, 13868265.

Salesforce Agentforce — salesforce.com/agentforce

Own key: no. Several assistants: yes — any number; Foundations is free to start. AI tools: yes — MCP via MuleSoft; “Salesforce – Beta” is in Anthropic’s directory. Your website: half — the final embed is a snippet, but it comes at the end of a multi-object build (channel, flow, deployment); no vendor badge documented; Salesforce’s docs say “We don’t support adding custom CSS.” Model: half — bring-your-own model through the Einstein Trust Layer. Sales call: half — free to start, but you need a Salesforce org. Reporting: yes — Agentforce Observability with Utterance Analysis (clusters of actual customer utterances, unsupported-request filter, per-agent usage); requires Data 360.

Sources: salesforce.com/agentforce/pricing · salesforce.com/agentforce/observability · Trailhead “Agentforce Analytics and Monitoring”.

Sierra — sierra.ai

Own key: no. Several assistants: no. AI tools: half — agents can be published as ChatGPT Apps; no Claude-facing server found. Your website: no — no public self-serve widget; deployment is sales-led and the documentation is gated. Model: no — vendor-chosen. Sales call: no — demo only; pricing not published. Reporting: yes — Insights and Explorer (natural-language questions across conversations, auto topic tagging, weekly briefings).

Sources: sierra.ai/product/explorer · sierra.ai/platform.

Tidio (Lyro) — tidio.com

Own key: no. Several assistants: no — one Lyro. AI tools: half — an install-and-embed helper connector; Lyro can call MCP tools. Your website: half — a one-line snippet, but the widget carries Tidio branding unless you buy the removal add-on ($16.67 a month) or the Plus plan; styling is a fixed option set with no CSS. Model: no — Claude, fixed. Sales call: yes — free tier, public pricing. Reporting: yes — Lyro Analytics: “Answers by intent” topics, a “Review unanswered” list of the actual questions it couldn’t answer, knowledge performance; available with the Lyro add-on on any plan. No digests or explicit period comparison.

Sources: tidio.com/pricing · help.tidio.com article 12642627073948 (Lyro Analytics).

Voiceflow — voiceflow.com

Own key: half — “bring your own model” on enterprise terms. Several assistants: half — 20 on Pro, unlimited on Business. AI tools: yes — a hosted OAuth MCP server to build, edit, publish and analyse agents from Claude Code and Cursor. Your website: half — snippet install and real custom CSS, but the Voiceflow watermark stays unless you’re on a Business plan (“Customers on a Business plan can disable Voiceflow branding”). Model: yes — OpenAI, Anthropic, Google, Groq. Sales call: half — trial available, but the public tier table has been withdrawn. Reporting: half — outcome and cost dashboard with previous-period comparison, transcripts and evaluations; no topic or gap reporting.

Sources: voiceflow.com/pricing · voiceflow.com/docs/documentation/measure/analytics · voiceflow.com/docs (MCP server).

Zendesk AI agents — zendesk.com

Own key: no. Several assistants: half — custom agents via Agent Builder, priced through seats. AI tools: half — an MCP server was announced in May 2026 for “early access this summer” and had not visibly shipped; an MCP client is in early access. Your website: half — snippet install, but the Zendesk logo sits on the widget and can be hidden only on Suite Enterprise; styling is theme options. Model: no. Sales call: yes — 14-day trial, public seat pricing. Reporting: half — contact-reason and resolution dashboards; discovered topics on Professional plans and above; no digests.

Sources: zendesk.com/pricing · zendesk.com/service/ai-agents · Zendesk help 9510024609178, 8357756417562.

copadre — copadre.com

Own key: yes — your own Anthropic key, stored encrypted; Anthropic bills you directly; no markup. Several assistants: yes — ten on the first tier, only active ones counted. AI tools: yes — connect Claude or any MCP-compatible tool to build, test and review, subject to your copadre permissions. Your website: yes — one line of code, and the widget carries no copadre badge, logo or link on any plan; it ships structure only and you style it entirely with your own CSS. (The only trace of us is the class names in the source.) Model: half — per-assistant choice from the current Claude range (Haiku, Sonnet, Opus); one vendor. Sales call: yes — three months free, no card. Reporting: yes — themes, your visitors’ own words, this month against last, daily and weekly digests, out-of-knowledge trends, volume and engagement by assistant.

Sources: copadre.com/pricing · copadre.com/features · copadre.com/terms.

What the table doesn’t measure

It doesn’t measure quality of answers, resolution rates, security certifications, service levels or support — the things enterprise buyers rightly ask about and that can’t be checked from a pricing page. It doesn’t measure price. It doesn’t measure messaging channels beyond your website — WhatsApp, Instagram, email, voice — where several platforms in this table are strong and copadre today is not; if your customers live in those channels, weigh that alongside this table. And it doesn’t measure the thing we think matters most, which is what happens when one assistant isn’t enough: how a platform lets several small, specialist assistants work as a team. That’s on our how it works page.

Tell us if we’ve got something wrong

Vendors change their plans, and we’d rather be accurate than flattering. If a score here is out of date, tell us through the contact page with a link to the page that shows it and we’ll re-check and update our scoring.

Last checked 28 August 2026.