SiteGPT Review 2026

How we evaluated this

We evaluated SiteGPT as a website chatbot with message quotas and optional inbox connectors, not as a full ticketing platform. Scoring weights source coverage, citation quality, quota honesty (including the double-count and model mix), sync cadence by plan, and handoff into an existing inbox.

We read the official homepage, pricing page, and security page on August 19, 2026. We kept the existing official product film already in the live HTML. We did not run a multi-week production pilot.

Pricing calculation to run

A "message" is both sides of the turn, and GPT-4.1 is about 10x mini

SiteGPT's public calculator is the most important object on the pricing page, and it is easy to ignore. Two rules change the Starter $39 story.

First: each billed message counts the user's question and the AI's reply. A 4,000-message Starter quota is about 2,000 conversational turns if every turn is one question and one answer. A visitor who asks three follow-ups has already consumed eight messages. Teams that forecast "4,000 chats" on Starter will blow the cap. Forecast turns, then multiply by two, then add a 20% buffer for clarifications.

Second: GPT-4.1-mini is described as about 10x cheaper than GPT-4.1, so the same plan's real capacity can vary by 10x. The calculator example on Growth yearly shows a 50/50 split producing 500 GPT-4.1 messages and 5,000 mini messages, totaling 5,500 against a 10,000-message plan. If you pin everything to GPT-4.1, Starter's 4,000 messages are a small production pool. If you pin everything to mini, you get more turns and, on many FAQ corpora, acceptable accuracy. Measure both on your 50-question set before you pick Scale.

• Starter: 1 chatbot, 4,000 messages, 1,000 pages, 1 member, manual refresh only, basic analytics. Fine for a quiet docs site. Wrong for a launch week.
• Growth: 2 chatbots, 10,000 messages, 10,000 pages, 4 members, API, monthly auto-refresh, advanced analytics, iOS app, weekly digest. This is the first collaboration plan.
• Scale: 3 chatbots, 40,000 messages, 50,000 pages, 10 members, weekly auto-refresh, daily auto-scan, webhooks, AI conversation insights, Zendesk AI Agent.
• Enterprise: up to 10,000 chatbots, 500,000 pages, 10,000 members, daily refresh, DPA, HIPAA-eligible, BAA available, priority support, custom integrations.
• Add-ons at $39 per month each ($468 yearly): remove SiteGPT branding, and extra 5,000 messages. Branding removal equals 100% of Starter's yearly-billing rate. Agencies should put it in the first quote, not the second invoice.

Worked month: a SaaS help center with 1,800 visitor sessions that open chat, average 2.5 turns (5 messages) each, mostly mini. That is 9,000 messages, which is Growth, not Starter. If 30% of turns use GPT-4.1 at a 10x rate, those 30% consume like 3x the mini volume and you are in Scale territory. Add $39 branding if the widget cannot show SiteGPT. The $39 Starter tile was never the quote. Files convert to pages at 2,500 cleaned characters per page, so a 50-page PDF can eat Starter's 1,000-page cap faster than a marketer expects.

Who SiteGPT fits best

Channel coverage to verify

The default is a website or help-center embed, plus a unique chatbot URL you can link to. Homepage copy says you can embed on as many sites as you want. Plan tiles still cap chatbot count (1 / 2 / 3 / 10,000), so "as many sites" is not "as many bots." Agencies should map chatbot caps to client count.

• Website and in-app embed: primary. Test authenticated app pages separately from the public marketing site.
• Crisp, Slack, Zendesk: named native integrations. Validate transcript passthrough, not only that the logo is on the page. Scale's "Zendesk AI Agent" line needs a live ticket test.
• iOS app: Growth and above.
• API: Growth+. Webhooks: Scale+.

WhatsApp and voice are not the public homepage lead. If you need those as first-class channels, compare Chatbase or YourGPT AI in the same pilot rather than assuming a SiteGPT plugin exists.

Workflow execution and handoff

The vendor's three steps are: sync training data, install on the site, then learn and refine from chat history and feedback. Higher plans add auto-refresh and a Monday digest of last week's numbers, topics, and unanswered questions. That digest is the operating loop. Without an owner, it is unread email.

Escalation is a user-initiated handoff to a live agent. Quality depends on the inbox you connect. A Crisp or Zendesk handoff that drops identity and the last five turns will make SiteGPT look worse than a form. Define what counts as a qualified lead before you enable capture, or you will fill a CRM with people who asked for the password-reset article.

• 24/7 FAQ deflection from docs and product pages.
• Lead capture on high-intent pages after the bot has answered a pricing or plan question.
• Zendesk-adjacent automation on Scale, after you watch a real ticket get created with context.

Pricing checked August 19, 2026

SiteGPT pricing: $39 to $259 per month billed yearly, Enterprise custom

The public page defaulted to yearly billing in our snapshot: Starter $39 per month billed $468 yearly, Growth $79 billed $948, Scale $259 billed $3,108. The toggle advertises Save 40%. The calculator also says Save 39% yearly. Pricing is exclusive of taxes. A 7-day free trial is offered. We did not capture a filled-in monthly-billing column, so do not treat $39 / $79 / $259 as proven month-to-month stickers.

Source: sitegpt.ai/pricing, August 19, 2026. Confirm the monthly toggle, message definition, model multipliers, and add-ons at checkout.

Add-ons

AI capability and grounding

Answers come from trained content with source context. Model choice changes effective capacity. The calculator names GPT-4.1 and GPT-4.1-mini. You can change the split after signup. Training usually finishes in minutes, depending on corpus size. Files larger than 10 MB are a contact-us path (bhanu@sitegpt.ai), not a self-serve upload.

Citations are the control. If an answer cannot open a source, treat it as untrusted. Structure help articles as short headed sections. A 40-page PDF converted at 2,500 characters per page will both burn the page cap and hide the clause retrieval needs.

Feature areas to verify

• URL, sitemap, file, YouTube, and connected-source training.
• Quick prompts to start common questions.
• Citations and chat-history feedback.
• Weekly email digest (Growth+).
• Lead capture and human escalation.
• Analytics, engagement funnel, AI topic grouping (higher plans).
• API (Growth+), webhooks (Scale+), Zendesk AI Agent (Scale+).
• Rate limiting (Growth+).
• 95+ languages.

Analytics and operating visibility

The homepage splits two views. Daily trends and an engagement funnel show whether people start, continue, and finish. AI-powered insights group conversations into topics and outcomes and surface repeated questions. Growth adds advanced analytics and a Monday digest. Scale adds AI conversation insights.

Use those views to patch source pages, not as a workforce suite. Keep helpdesk CSAT and repeat-contact in the system of record. Baseline week one so a "the bot is working" slide has a denominator. Export unanswered questions. That list is the content backlog.

Security, data handling, and compliance

The security page (August 19, 2026) states: SOC 2 Type II examination covering security, availability, and confidentiality, audited with zero exceptions; GDPR certified by DPLMC International; HIPAA assessed by DPLMC International; conversations and training data never used to train or fine-tune models; zero data retention with LLM providers per OpenAI API policy as cited; export or delete anytime; TLS 1.2+ in transit and encryption at rest; role-based access; Cloudflare Workers plus Prisma Accelerate; published subprocessors.

• SOC 2 report is available for Enterprise customers via trust.sitegpt.ai. Request it in the security review, not after signature.
• DPA is available. The security page says DPA for Enterprise customers in one block and also links a DPA. Confirm which plans get a signed DPA.
• BAA and HIPAA eligibility are Enterprise plan items on the pricing tile.
• Infrastructure on Cloudflare Workers is a concrete choice. Ask how conversation data is stored and where.

What questions should you ask before buying SiteGPT?

• What is our peak-week message count if each turn is two messages and 20% of turns use GPT-4.1?
• Which sources must auto-sync, and is monthly (Growth) enough versus weekly (Scale)?
• Show a Zendesk or Crisp escalation with transcript and identity intact.
• Is the $39 / $79 / $259 figure yearly-billing only, and what is month-to-month?
• Do we need branding removal and extra message packs in month one?
• Can we upload our 10 MB+ policy PDFs without a custom arrangement?
• Who owns the Monday digest and the unanswered-question list?
• Do we need a signed DPA or BAA on day one, which means Enterprise?

What red flags should you watch for with SiteGPT?

• No peak-volume model that includes the double-count and model mix.
• Starter sold for a site that will exceed 4,000 messages after follow-ups.
• No escalation test into the real inbox.
• Set-and-forget embed with no content owner. Manual refresh on Starter will go stale.
• Agency quotes that omit $39 branding per client.
• Assuming deep order actions work because Zendesk is on the Scale tile.

What are the best alternatives to SiteGPT?

Pick by meter (messages versus credits versus resolutions) and by the inbox you already run.

• ChatbaseChoose Chatbase when you need procedures, optional voice, and a credit-plus-add-on ladder rather than SiteGPT's message quota.
• YourGPT AIChoose YourGPT when omnichannel actions across WhatsApp, email, and web are the job.
• Intercom FinChoose Fin when the team already lives in Intercom and you want resolution-priced AI on that inbox.
• Dante AIChoose Dante when you prefer a flat fee plus a published per-model credit ladder, and you can live with handover as a Pro feature or a pack.
• TidioChoose Tidio when live chat plus a lighter bot is enough and you do not need SiteGPT's source connectors.

Workflow test

What SiteGPT needs to prove in a real workflow

1. Connect the core site and the documents that drive tickets.Score 50 real questions with citations open. Count grounded correct, not "bot answered."
2. Measure a full business week of messageswith your intended GPT-4.1 / mini split. Remember: question plus reply equals two. Compare that week to Starter 4,000 versus Growth 10,000 versus Scale 40,000.
3. Escalate end to endinto the inbox you will actually staff. The human must see the thread. If they do not, the hybrid model is a slogan.
4. Model Growth versus Scale with peak and add-onsincluding $39 branding and extra 5k packs. Then choose billing term after you have seen the monthly toggle, not only the yearly default.

• Chatbase review
• Dante AI review
• YourGPT AI review
• Intercom Fin review
• Tidio review
• Customer support AI agents buyer guide
• How to choose an AI agent platform

Official product demo

See SiteGPT before you trust the claims

This official SiteGPT introduction is a product tour. It is not independent proof of quota math, citation quality, or implementation effort. After watching, repeat the workflow with your own pages, files, and a forced escalation.

View on YouTube and ask the vendor to repeat the workflow with your own content and failure cases.

Claim and source ledger

Should you choose SiteGPT?

Choose SiteGPT when you want a content-grounded website agent with a public message ladder, citations, and a 7-day trial, and when someone will own the unanswered-question list. The August 19, 2026 yearly-billing card is $39 / $79 / $259 plus custom Enterprise, then $39 add-ons for branding and extra 5,000 messages. Count both sides of every turn. Split quota between GPT-4.1 and mini on purpose.

Do not make it the only system if you need a full enterprise contact center or unproven authenticated actions. Confirm the monthly toggle before you assume the yearly-default dollars are also month-to-month. Run the four-step workflow test, including a peak week, before annual billing.