To check whether an AI assistant recommends your app, start with a small set of questions that describe the jobs your app does. Save the answers, check the app identities and cited pages, and keep the conditions of each check. That gives you something you can inspect again later.
You can do this with a note and the AI assistant you already use. The worksheet records the answers you observe. Other people may receive different answers.
I make AppFound, a NaguBuilds project. I'm building it for app founders who want to understand how their apps appear in AI recommendations. It is currently prelaunch. This guide gives you a way to begin the research yourself.
Start with the question a customer would ask
Write one sentence about the person and the task before opening an AI assistant. For example, a founder of a subscription tracker could investigate this question:
Which iPhone apps let me track subscriptions without connecting my bank account?
This is an illustrative research question, not a recorded answer or a finding about a particular app. It contains a task, a platform and a constraint that could change the shortlist.
Use your actual features to write several questions. Start with the ordinary task, then try a constraint that would change the choice of app. Avoid adding your brand to every question. Asking an assistant to describe an app you named tests a different situation from asking it to suggest an app.
Keep the questions you choose, including the ones where you expect your app to struggle. Replacing them after seeing disappointing answers makes a later comparison difficult to interpret.
Record the conditions before collecting an answer
Give each question an ID. Record the assistant and visible mode, the date, language, region used in the question, and whether the conversation starts fresh. If account personalization is active, record that too. You do not need to publish your account details.
Use the same conditions for a later check where you can. If something changes, preserve that change beside the result instead of presenting it as a clean before-and-after comparison.
For ChatGPT, record whether the answer used web search and keep its linked sources. OpenAI explains how citations and the sources panel work in its ChatGPT search guide. An answer without visible sources still tells you what appeared in that conversation; it gives you less material for checking where particular claims came from.
Separate what appeared from what you think it means
Read the whole answer before marking your app as recommended. Use these working definitions:
- A mention means the answer names the app.
- A recommendation means the answer presents the app as an option for the person's task.
- A citation means the answer links to a page. The cited page may concern another app or a general topic.
These are definitions for this worksheet. A tool may use different definitions, which is why its raw evidence matters.
An app might appear only in a warning, a comparison, or a sentence saying it does not meet the requirement. Record that context. Do not count a warning as a positive recommendation just because the brand appears.
Next, check identity. Follow the official store or developer link and confirm the app name, developer, platform and store ID. Apple describes how country and region availability works. A name match alone can hide the wrong app or an unavailable version.
Copy this observation worksheet
Create one record per question and answer. Leave unknown fields unknown.
- Question ID and exact wording:
- Date and time:
- Assistant, visible mode and language:
- Region in the question and relevant storefront:
- Fresh conversation or existing context:
- Search used, not used, or unclear:
- App name, developer and verified store URL:
- Exact short passage about the app:
- Mention, recommendation, citation, or absent:
- Linked source and the claim it supports:
- Feature, price or availability claim checked:
- Incorrect or unsupported claim:
- Missing evidence or interrupted collection:
- One possible follow-up, with its reason:
You can copy this list into a note or spreadsheet. Save enough of the original answer to check your classification, while keeping private conversation details out of anything you share.
Check the cited page, then choose a follow-up
Open each source that matters to the recommendation. Does it concern the right app? Does it support the claimed feature? Is the price current in the relevant storefront? If the source no longer loads, preserve the URL and mark the claim unresolved.
OpenAI acknowledges that ChatGPT can produce incorrect information and fabricated references, so the answer itself is not final proof of a product fact. Its accuracy guidance recommends checking important information against reliable sources.
Choose a follow-up you can explain from the record. If your own public page omits a feature your app really has, clarifying that page helps readers understand the product. If a cited third-party page is wrong, record the exact error before deciding how to address it. Sometimes the record supports no change yet. Keep it and check again.
Avoid asking the assistant why it omitted your app and treating the explanation as access to its hidden reasoning. Keep your interpretation separate from the saved answer.
Repeat without turning a sample into a promise
Repeat the same questions on another occasion if your access allows it. Compare the actual records. Note changes to the assistant, question set or product pages, and include failed checks rather than quietly dropping them.
If you summarize results, state the numerator, denominator and conditions: how many usable answers included a recommendation, out of how many usable answers you collected. A few observations do not establish market share, a universal rank, or the effect of an edit.
AppFound is being built around these questions. You can visit AppFound and join its prelaunch waitlist; the report on its current homepage is labelled as an example. The worksheet above is ready to use without it.