Discontinued products in AI answers: end the offer, keep the history
Ending sales of a product does not remove the instructions, comparisons and references that explain what it was. Removing everything at once can deprive existing customers of useful documentation. Changing nothing except the buy button can leave an old page presenting a discontinued offer as current.
This procedure is for product, ecommerce and support teams after a retirement decision has been approved. The deliverable is a record separating sales, use and historical documentation, with a decision for each channel and follow-up checks of AI recommendations. It does not require erasing every mention or promise that an assistant will update within a particular period.
Tracking product recommendations at SKU level can reveal the problem. This procedure starts after detection: closing the offer without confusing prospective buyers with people who need to maintain the product.
Confirm what ends and when
Do not turn a temporary stock shortage into permanent retirement. Before changing pages or feeds, product owners must confirm the status, affected variants and markets. One generation can stop selling in one country while remaining available elsewhere. A brand can also end direct sales while distributors still have stock.
Record these decisions separately:
| Decision | What must be clear |
|---|---|
| End of direct sales | The date from which new orders are no longer accepted |
| Remaining third-party stock | Which sellers and markets were checked, and when |
| Continuing support | Which documentation and services remain available |
| Commercial successor | Which product is proposed and its validated uses |
| Historical reference | Which statements describe an earlier period |
Do not infer the end of support from the end of sales. Nor should a product sharing the same category be described as a compatible successor without evidence. Keep unapproved dates and conditions marked as pending rather than filling the record with plausible assumptions.
Open a retirement record for the offer
Use the stable catalog identifier and link previous names, variants and URLs. The AI brand entity audit supplies those relationships; there is no need to build another complete identity inventory.
The record should contain the approved decision, effective date, affected markets, confirmation source, owner and exceptions. Add separate statements about current sales, support and succession where relevant. Each needs its own evidence.
Keep a change log with the URL or channel, previous text, planned action, owner, implementation date and evidence of the result. An item marked “updated” does not establish whether the storefront, PDF or only the internal document changed.
Retain the previous version under appropriate access controls. It may help resolve a documentation dispute or interpret an old citation without presenting it as the current commercial offer. Retention periods and permissions depend on your organization's applicable policies.
Choose a useful archive, a successor or removal
The decision depends on the purpose of the URL. A page with manuals, compatibility information and spare parts can remain useful after sales end. If retained, it should state prominently that the specified sales offer has ended, show the date and distinguish support from any commercial alternative.
A redirect to a successor makes sense only when the destination genuinely addresses the old page's purpose. Document relevant differences rather than hiding them behind “new model.” If maintenance needs remain that the successor cannot serve, preserve an accessible documentation destination.
Google distinguishes a URL with a clear replacement from a deleted page without one: a permanent redirect can serve the former, while 404 or 410 responses apply to the latter. This is not a rule to redirect an entire retired catalog to the homepage. Google's crawling error documentation.
For each URL, record the reason for the decision and test the result. A page saying “not found” with a 200 response is not equivalent to proper removal. A useful historical page should not be treated as missing merely because it no longer accepts orders.
Separate the web archive from the commercial catalog
Review visible text, purchasing controls, structured data and feeds as distinct surfaces. Removing a button does not fix a description promising immediate delivery. Keeping a manual does not require keeping a commercial offer active.
Google Merchant Center's documentation says to remove permanently discontinued products from the data source rather than retain them as merely out of stock. That rule applies to that channel and should not automatically be copied to every recipient. Merchant Center availability rules.
Schema.org provides the Discontinued value. It describes a status, not how every platform will interpret it, and does not replace ingestion rules. Avoid markup that contradicts what a buyer can actually do on the page.
For each feed, verify the current mechanism for removing an offer and retain its processing result. The ChatGPT product feed guide covers delivery contracts. Do not assume that leaving a row out of a partial upload deletes it from the destination.
Synchronize channels still presenting the offer
Start with the URLs and documents in the record, not a campaign to change every mention on the internet. Review category pages, your own comparison pages, internal search, PDF catalogs, help content, sales messages and relevant distributor listings.
Assign an action according to your level of control:
- On owned surfaces, publish the approved status and check the publicly accessible version.
- For integration-managed channels, confirm receipt and final status rather than transmission alone.
- On third-party pages, record the contradiction and request a correction where appropriate. Do not mark the change complete until it can be checked.
- Preserve the time context of accurate historical references. An old review need not become a current sales page.
Downloadable documents need visible version control. If an earlier catalog must remain available, its date and scope should distinguish it from the current catalog. Prioritize contradictions that could lead to an impossible purchase or an incompatible substitution.
Review buying, support and historical recommendations separately
A mention of the retired product can be correct. Evaluation should depend on the question and the offer that actually exists in that market, not on the appearance of an old name.
| Query intent | What to check |
|---|---|
| Buying now | Do not present a closed offer as available |
| Finding an alternative | Distinguish a verified successor from a similar option |
| Maintenance or compatibility | Preserve the old product's identity and documentation |
| Historical research | Place facts and availability in their time context |
| Buying secondhand | Do not confuse resale with current direct sales |
Prepare questions using the names and markets in the record without supplying the expected answer in every question. Keep unprompted category questions too: if every query starts by saying the model is discontinued, the test only shows what happens after the assistant receives that information.
Save the question, full response, date, configuration, sources and problematic passage. Check the commercial destination directly when one exists. An old citation alone does not establish that the assistant is recommending an invalid purchase.
Verify retirement without claiming an unproven cause
Maintain two statuses: completion of retirement actions in your channels and behavior observed in answers. The first can be resolved while the second remains pending. Do not reopen an offer to make a measurement issue disappear.
For follow-up checks, use comparable conditions and multiple observations. Classify confirmed invalid recommendations, correct historical information, ambiguous responses and non-evaluable executions. If the destination is a distributor, check its current offer before labeling the result an error.
Report counts and denominators: valid current-purchase responses containing an invalid recommendation divided by valid current-purchase responses, separated by market and surface. Do not mix in support questions to make the rate look lower.
An error disappearing after a source update does not prove that the edit caused the change. Record what changed and when, what was observed afterward and what remains unattributable. There is no universal assistant-update deadline that can be inferred from closing a feed entry.
Close with visible exceptions and a next review
Closure requires evidence that the direct offer is disabled where appropriate, chosen destinations work, support terms are coherent and outstanding external channels have owners. Attach subsequent response checks and their limitations.
Do not demand zero product mentions. Require known material errors to be corrected or covered by an explicit exception, with someone responsible for remaining work. A distributor with verified stock may continue selling, while a historical manual may remain the best answer for an existing user.
Schedule the next review around remaining risk and decisions that can still change, not a supposed date when the model will forget. If you use Mentio to observe responses, link those observations to the retirement record. This procedure does not announce automatic retirement, feed synchronization or knowledge deletion features in the product.
Frequently asked questions
Do discontinued and out of stock mean the same thing? No. A stock shortage can be temporary; permanent retirement closes a specific offer from a given date. Confirm variants, market, seller and continuing support before changing pages or channels.
Should a discontinued product page be deleted? Not always. It may still serve maintenance or historical documentation needs. Decide whether to retain a useful archive, redirect to an appropriate replacement or remove a page without a replacement, then test the chosen technical result.
Is changing availability in structured data enough? No. Visible text, purchasing controls, markup and each feed must reflect the actual situation. Recipients have their own rules; changing markup does not confirm that an offer has been removed from their catalogs.
Is an AI answer always wrong when it mentions the old product? No. It can correctly answer a historical, support or secondhand query. Assess an error against the intent, market, seller and date of the offer it recommends.
When can product retirement be considered complete? When owned actions are verified and exceptions have an owner, evidence and a planned review. Commercial closure and answer monitoring are separate statuses; there is no need to erase every historical reference or promise that assistants will update immediately.
Want to know if AI mentions your brand?
Discover your visibility in ChatGPT, Claude and Gemini in minutes.
Related articles
How to Track AI Product Recommendations at SKU Level
Track SKU inclusion, position, attributes, availability and substitutes across ChatGPT, Gemini and Perplexity with a comparable protocol.
AI CommerceChatGPT Product Feed: How Merchants Build and Maintain It
Build, validate, and maintain a ChatGPT product feed with required fields, variants, freshness, eligibility controls, and operational QA.
GEO StrategyAI Brand Entity Audit: Names, Products, Aliases and Facts
Build a verifiable map of brand names, products, aliases and facts to find ambiguity before optimizing your brand's visibility in AI answers.