CustomGPT - Agent

The problem with putting a general chatbot on your website is that it answers confidently from general knowledge rather than from your documentation. CustomGPT addresses that by constraining answers to content you supply and attaching source citations to each response. That grounding is the entire value proposition, and the price reflects a business tool rather than a hobbyist one.

What It Actually Does

You point the platform at your content: a website, a sitemap, uploaded documents, or connected systems. It crawls, chunks, and indexes that material. A retrieval layer then finds relevant passages for each question and instructs the model to answer only from those passages.

The result deploys as a chat widget on your own domain, through an API, or into workplace tools. End users need no account with anyone. Every answer can display the sources it drew from.

Key Features

• Citation-backed answers — Responses link to source material, which lets users verify claims and lets you audit failures.
• Sitemap ingestion — Point it at a sitemap and content updates flow in without manual re-uploading.
• Broad connector support — Integrations with document stores, CRMs, and workplace tools consolidate scattered knowledge.
• REST API on paid tiers — Every paying plan reportedly exposes an API, which many competitors reserve for enterprise.
• Multilingual responses — The bot answers across many languages from the same underlying corpus.
• Compliance posture — Recognised security certification and a stated policy of not training on customer data.

Where It Fits Best

Mid-market customer support and knowledge operations teams are the natural buyer. The economics work when repetitive questions consume real agent time. Reviewers suggest the subscription is hard to justify below roughly fifty questions weekly.

Regulated and document-heavy sectors appear frequently in its customer base, including legal, education, and government. Those buyers value citations and data handling more than they value the lowest price.

Tradeoffs Worth Knowing

Cost is the most consistent criticism across review sites. Entry pricing is reported near $99 monthly, with a premium tier around $499. Multiple reviewers state plainly that comparable feature sets exist for less, and at least one describes cancelling after concluding it was more expensive than alternatives doing the same job.

Architectural limits matter if you expected a general agent platform. There is no tool-use loop you control, no scripting layer, and no choice of underlying model. It builds retrieval chatbots well and does not attempt to be a framework for autonomous agents.

Self-hosting is not available. For organisations whose compliance requirements demand data stay on their own infrastructure, that rules it out regardless of certification.

Public sentiment is genuinely mixed rather than uniformly positive. Alongside praise for accuracy and support, at least one public reviewer described the product as not ready for public use. Treat the marketing accuracy benchmarks with corresponding caution, since those figures are published by the vendor itself.

Practical Notes

Citations change the support dynamic in a useful way. Users can check the source themselves, which reduces disputes about whether the bot invented an answer.

Compare it against building a custom GPT inside a consumer chat product before buying. That route is cheaper and works fine for a personal assistant. It fails when you need a public-facing bot on your own domain with an API, which is exactly the gap this fills.

Deployment is genuinely quick. Reviewers describe ingesting a site and having a working bot within roughly half an hour, which is unusual for this category.

Plan ownership before purchase. This works best where one or two people own the bot project end to end, maintaining the corpus and reviewing failed answers. Without that ownership the content drifts stale and answer quality degrades quietly.

Content hygiene drives answer quality more than any setting. Contradictory or outdated documentation produces contradictory answers, and the citation feature makes that visible to your customers.

Watch how the bot handles questions outside its corpus. A grounded system should decline rather than improvise. Test that behaviour deliberately, since it is where reputational risk actually sits.

How It Compares

Against general model APIs, you are paying for a managed retrieval pipeline rather than raw capability. That is a reasonable trade if you do not want to build and maintain chunking, embedding, and retrieval yourself. Against cheaper chatbot builders, the differentiators are citations, compliance documentation, and API availability across tiers. Against self-hosted open-source retrieval stacks, you give up control and gain not having to run anything.

What to Verify Before Choosing CustomGPT

• Query volume included at your tier and the cost of exceeding it
• Whether your ticket volume justifies the subscription against agent time saved
• Answer accuracy tested on your own awkward edge-case questions
• Whether self-hosting is a hard requirement for your compliance team
• Storage and item limits against the size of your documentation
• Which connectors cover your actual content sources
• Total cost compared against cheaper builders with similar features
• How re-crawling works when your documentation changes
• Whether answer logs are available for reviewing failed queries