A Product Compatibility Chatbot That Does Not Guess the Fit
“Will this replacement fit the one I already own?” is a purchase question, not a request for a product description. A visitor may know the brand but not the exact model. Your store may sell several versions with nearly identical names. A confident answer based on a shared word can send the wrong item to checkout.
A source-based product compatibility chatbot should explain the fit your store actually documents, identify missing model information, and direct unresolved cases to the right support route. This guide shows how to prepare that setup in Simple Agent. It does not assume a connection to a live catalog or a tool that inspects the customer's equipment.
Put the exact supported models beside the product
Start with one product that regularly attracts compatibility questions. Read its public page as someone holding an older version of the item would. Can that person find the full product name, the replacement reference, the supported model identifiers, and any excluded versions?
Do not use a broad category such as “fits our desktop organizers” when only a particular range is supported. Similar dimensions, a shared brand, or a matching photograph do not establish compatibility. If the manufacturer provides a compatibility list, check that the list covers the exact item you sell and link to the relevant source. Do not extend that list by inference.
For illustration, imagine a fictional stationery store selling replacement trays for modular desk organizers. Its approved product information might say:
The Slim Tray ST-2 fits the Desk Stack D2 organizer. It does not fit Desk Stack D1. Find the model code on the underside of your organizer. If the label is missing or you have a different model, contact our product support team through the store's Contact page before ordering.
These names and rules are fictional, not claims about real products. Your source needs your actual model identifiers and verified fit information. A short, explicit exclusion is more useful than a paragraph about versatility.
Train on the product facts, not the whole sales archive
The Simple Agent getting-started guide describes training sources supplied through public URLs, PDFs, and direct text. Training on your sources and using the website widget require an active paid plan. In that workspace, begin with the chosen product page and the current compatibility guide.
If the distinction between versions is buried in a long manual, a direct-text FAQ can make it clearer. Have the person responsible for the product verify the wording first. The FAQ should agree with the product page, rather than quietly introduce a second set of rules.
Old launch announcements and retired product pages deserve particular attention. A statement that a tray fits “all current organizers” can change meaning as the catalog grows. Replace relative wording with explicit identifiers where possible. Resolve contradictions before adding both documents as sources; do not expect the agent to decide which commercial promise your store intends to honor.
You do not need customer order histories to explain a published fit rule. Keep private purchase information out of the training material for this task. Knowing the compatibility policy is different from knowing which version a customer bought.
Check near-matches, missing details, and unsupported versions
Use the Agent Builder's Playground to ask questions based on the exact source. Include the straightforward case, but spend more attention on questions where a wrong answer would still sound plausible.
For the fictional tray above, check these questions separately:
- “Does ST-2 fit Desk Stack D2?” The reply should explain the documented fit and cite the relevant source.
- “Does it fit D1 as well?” The reply should preserve the explicit exclusion, not turn it into “probably.”
- “I have a Desk Stack, but I don't know the version.” The reply should explain where the model code is located and provide the existing support route if needed.
- “Mine is D2 Plus. Is that close enough?” The reply should not treat a similar name as a supported model when the source does not list it.
Also ask a follow-up after the supported case: “So it will fit my older one too?” The answer must keep the model distinction. Agreement in one turn should not become an assurance about an unspecified product in the next.
If a reply overstates fit, compare it with the exact passage supplied. Correct ambiguous or contradictory source material, update the source in the workspace, and repeat the questions. Do not approve a response just because it is polite or includes a citation: open that citation and check whether it supports the specific model claim.
Keep a product link separate from a fit guarantee
Once the replies match the source, follow the Embed instructions in the getting-started guide to install the widget. Ask the same questions on the published website and follow the product and support links on a phone as well as a larger screen.
A link to a product page should help the visitor inspect the right item. It should not imply that compatibility has been established for an unlisted version. Likewise, sending someone to Contact is not a promise that a replacement has been approved, reserved, or ordered.
For unresolved cases, use the contact process your store already operates. Do not invent a dedicated compatibility team, a response deadline, or a photo-upload service if those are not available. If the visitor needs a model code, explain how to find it using your verified instructions. The objective is a usable next step, not a vague request to “check the details.”
Choose source-based answers or a connected catalog deliberately
This setup is suited to published product facts and stable compatibility rules. If your requirement is live inventory, personalized order lookup, or compatibility calculated from a connected catalog, evaluate that connection separately. Uploading a product guide does not create access to those systems.
Recheck the questions when you add a model, change a replacement reference, or retire a product. Keep the public page and the agent's sources aligned before promoting the new range.
To implement the source-based workflow, review Simple Agent's paid plans, then prepare one product page with explicit supported and unsupported versions. The useful outcome is not an automatic “yes.” It is an answer that helps the shopper choose the documented fit—or reach your real support process before ordering the wrong item.
Ready to build your AI agent?
Guided setup in 90 seconds. One flat price, published monthly volume.
Create my agent