Author
Patrick Bass Submitted
Sep 5, 2026 4:34am Summary
Family meal sharing tickets: the conditions behind the share button and the Share with family? switch, reading a "logged but not shared" report back to a member, sharing a saved meal without logging it, and what to escalate
'''What support needs to know about the mobieusAI second opinion.'''
== It is not the interaction alerts, and that is the first thing to establish ==
A member writing in about "the interaction check" could mean either of two features on two different pages. Ask which page they were on.
* '''Interaction alerts''' (Health → Medications, and /health/interactions) is a lookup against FDA product labeling. Quotes the label sentence, the section and the effective date. Deterministic: the same list gives the same answer. '''Free.'''
* '''The mobieusAI second opinion''' (Health → Supplements and Meds, at the bottom of the page) is generated. It reaches combinations no label states, and cannot cite a source. '''Requires mobieusAI to be on for the workspace, and the Supplements page is Premium.'''
If a member reports that "the check says something different than last time", establish which one first. Different answers on repeat runs are '''expected''' for the second opinion and would be a defect on the alerts.
== "I don't have the mobieusAI panel" ==
The panel renders only when all of these are true. Work down the list:
1. The tenant is on '''pro, creator, creator-plus or sovereign'''.
1. '''AI_FEATURES_ENABLED''' is 1 in the tenant environment.
1. The '''ai_health_supplement_check''' feature flag is on for that tenant in the admin plane.
1. The tenant has a '''mobieusAI key''' configured.
1. The monthly spend cap and the member's daily call cap have not been hit.
A member who has run it before keeps seeing their last answer even if the feature is later switched off. That is deliberate, not a bug. The result is theirs.
== "It asked me to look at what it sends" ==
Working as designed. Before the first run, the member is shown the disclosure page listing exactly what would be sent, and can decline. '''Not now''' declines this occasion only and does not disable anything.
What it sends: every supplement (name, dose, timing), every medication on their list (name, strength, frequency, timing), and from Health Context: prescription meds, conditions, pregnant/nursing, age, sex.
What is kept: the answer. The payload is not kept.
A member can turn the feature off permanently under '''Your Data'''. If they do, the panel answers with an opted-out message rather than silently doing nothing.
== "It flagged something alarming" ==
Do not interpret it, and do not reassure. The correct response is the one the feature itself gives: '''take it to a pharmacist or prescriber.'''
Points worth making to a worried member:
* Severity here is the model's reading, '''not''' where a drug label puts the information. On the alerts, severity does mean exactly that, which is another reason to establish which feature they are looking at.
* It never says a combination is safe, and nothing being flagged is not an all-clear.
* It never gives a dose and never tells anyone to stop a medication. If a member says it told them to stop something, get a screenshot and escalate. That is outside what the feature is built to produce.
== "It said my medication is spelled wrong" ==
That is a real and useful finding, and worth encouraging them to act on. A name Loopa cannot match to the reference vocabulary '''is excluded from the interaction alerts entirely'''. The alert page says how many items were left out, and a typo is the usual cause. Correcting the spelling on their list puts that item back into the label-based check.
== "It is unavailable right now" ==
The member sees this when the model call fails: no tenant key, a cap reached, or an upstream timeout. It does '''not''' mean their interaction alerts are down. Those are a different route and a local lookup, and still answer.
Check, in order: the tenant key, then the monthly cap and the member's daily count in /admin/ai, then the tenant's AI spend log.
== What to escalate ==
* A member reporting advice to start, stop, or change a medication or a dose.
* A flag naming an item the member does not have on their list.
* The panel appearing for a member whose workspace has the feature switched off.
+
* Any report that the two checks have been merged into one on screen. They must remain visibly separate, with the generated one labelled as generated.
−
* Any report that the two checks have been merged into one on screen. They must remain visibly separate, with the generated one labeled as generated.
[[Category:Support]]
[[Category:mobieusHealth]]
'''What support needs to know about the mobieusAI second opinion.'''
== It is not the interaction alerts, and that is the first thing to establish ==
A member writing in about "the interaction check" could mean either of two features on two different pages. Ask which page they were on.
- '''Interaction alerts''' (Health → Medications, and /health/interactions) is a lookup against FDA product labeling. Quotes the label sentence, the section and the effective date. Deterministic: the same list gives the same answer. '''Free.'''
- '''The mobieusAI second opinion''' (Health → Supplements and Meds, at the bottom of the page) is generated. It reaches combinations no label states, and cannot cite a source. '''Requires mobieusAI to be on for the workspace, and the Supplements page is Premium.'''
If a member reports that "the check says something different than last time", establish which one first. Different answers on repeat runs are '''expected''' for the second opinion and would be a defect on the alerts.
== "I don't have the mobieusAI panel" ==
The panel renders only when all of these are true. Work down the list:
- The tenant is on '''pro, creator, creator-plus or sovereign'''.
- '''AI_FEATURES_ENABLED''' is 1 in the tenant environment.
- The '''ai_health_supplement_check''' feature flag is on for that tenant in the admin plane.
- The tenant has a '''mobieusAI key''' configured.
- The monthly spend cap and the member's daily call cap have not been hit.
A member who has run it before keeps seeing their last answer even if the feature is later switched off. That is deliberate, not a bug. The result is theirs.
== "It asked me to look at what it sends" ==
Working as designed. Before the first run, the member is shown the disclosure page listing exactly what would be sent, and can decline. '''Not now''' declines this occasion only and does not disable anything.
What it sends: every supplement (name, dose, timing), every medication on their list (name, strength, frequency, timing), and from Health Context: prescription meds, conditions, pregnant/nursing, age, sex.
What is kept: the answer. The payload is not kept.
A member can turn the feature off permanently under '''Your Data'''. If they do, the panel answers with an opted-out message rather than silently doing nothing.
== "It flagged something alarming" ==
Do not interpret it, and do not reassure. The correct response is the one the feature itself gives: '''take it to a pharmacist or prescriber.'''
Points worth making to a worried member:
- Severity here is the model's reading, '''not''' where a drug label puts the information. On the alerts, severity does mean exactly that, which is another reason to establish which feature they are looking at.
- It never says a combination is safe, and nothing being flagged is not an all-clear.
- It never gives a dose and never tells anyone to stop a medication. If a member says it told them to stop something, get a screenshot and escalate. That is outside what the feature is built to produce.
== "It said my medication is spelled wrong" ==
That is a real and useful finding, and worth encouraging them to act on. A name Loopa cannot match to the reference vocabulary '''is excluded from the interaction alerts entirely'''. The alert page says how many items were left out, and a typo is the usual cause. Correcting the spelling on their list puts that item back into the label-based check.
== "It is unavailable right now" ==
The member sees this when the model call fails: no tenant key, a cap reached, or an upstream timeout. It does '''not''' mean their interaction alerts are down. Those are a different route and a local lookup, and still answer.
Check, in order: the tenant key, then the monthly cap and the member's daily count in /admin/ai, then the tenant's AI spend log.
== What to escalate ==
- A member reporting advice to start, stop, or change a medication or a dose.
- A flag naming an item the member does not have on their list.
- The panel appearing for a member whose workspace has the feature switched off.
- Any report that the two checks have been merged into one on screen. They must remain visibly separate, with the generated one labelled as generated.
[[Category:Support]] [[Category:mobieusHealth]]
'''What support needs to know about the mobieusAI second opinion.''' == It is not the interaction alerts, and that is the first thing to establish == A member writing in about "the interaction check" could mean either of two features on two different pages. Ask which page they were on. * '''Interaction alerts''' (Health → Medications, and /health/interactions) is a lookup against FDA product labeling. Quotes the label sentence, the section and the effective date. Deterministic: the same list gives the same answer. '''Free.''' * '''The mobieusAI second opinion''' (Health → Supplements and Meds, at the bottom of the page) is generated. It reaches combinations no label states, and cannot cite a source. '''Requires mobieusAI to be on for the workspace, and the Supplements page is Premium.''' If a member reports that "the check says something different than last time", establish which one first. Different answers on repeat runs are '''expected''' for the second opinion and would be a defect on the alerts. == "I don't have the mobieusAI panel" == The panel renders only when all of these are true. Work down the list: 1. The tenant is on '''pro, creator, creator-plus or sovereign'''. 1. '''AI_FEATURES_ENABLED''' is 1 in the tenant environment. 1. The '''ai_health_supplement_check''' feature flag is on for that tenant in the admin plane. 1. The tenant has a '''mobieusAI key''' configured. 1. The monthly spend cap and the member's daily call cap have not been hit. A member who has run it before keeps seeing their last answer even if the feature is later switched off. That is deliberate, not a bug. The result is theirs. == "It asked me to look at what it sends" == Working as designed. Before the first run, the member is shown the disclosure page listing exactly what would be sent, and can decline. '''Not now''' declines this occasion only and does not disable anything. What it sends: every supplement (name, dose, timing), every medication on their list (name, strength, frequency, timing), and from Health Context: prescription meds, conditions, pregnant/nursing, age, sex. What is kept: the answer. The payload is not kept. A member can turn the feature off permanently under '''Your Data'''. If they do, the panel answers with an opted-out message rather than silently doing nothing. == "It flagged something alarming" == Do not interpret it, and do not reassure. The correct response is the one the feature itself gives: '''take it to a pharmacist or prescriber.''' Points worth making to a worried member: * Severity here is the model's reading, '''not''' where a drug label puts the information. On the alerts, severity does mean exactly that, which is another reason to establish which feature they are looking at. * It never says a combination is safe, and nothing being flagged is not an all-clear. * It never gives a dose and never tells anyone to stop a medication. If a member says it told them to stop something, get a screenshot and escalate. That is outside what the feature is built to produce. == "It said my medication is spelled wrong" == That is a real and useful finding, and worth encouraging them to act on. A name Loopa cannot match to the reference vocabulary '''is excluded from the interaction alerts entirely'''. The alert page says how many items were left out, and a typo is the usual cause. Correcting the spelling on their list puts that item back into the label-based check. == "It is unavailable right now" == The member sees this when the model call fails: no tenant key, a cap reached, or an upstream timeout. It does '''not''' mean their interaction alerts are down. Those are a different route and a local lookup, and still answer. Check, in order: the tenant key, then the monthly cap and the member's daily count in /admin/ai, then the tenant's AI spend log. == What to escalate == * A member reporting advice to start, stop, or change a medication or a dose. * A flag naming an item the member does not have on their list. * The panel appearing for a member whose workspace has the feature switched off. * Any report that the two checks have been merged into one on screen. They must remain visibly separate, with the generated one labelled as generated. [[Category:Support]] [[Category:mobieusHealth]]
