AI citations
Who do AI assistants cite when you ask them to debug production?
We ask Perplexity and Gemini the same 50 production debugging questions, from CrashLoopBackOff to Postgres replication lag to a Cloudflare 522, and record every source they cite. The question set has been frozen since 2026-09-30 so weeks compare. We publish it as a dataset: anyone writing for engineers can see which sites the assistants actually read. We have a stake, we want aurorasre.ai cited, and our count is in the small print under each card.
Perplexityperplexity/sonar
50 of 50 answered, 370 citations across 229 domains, 2026-10-21.
- 1.github.com23
- 2.docs.aws.amazon.com17
- 3.docs.cloud.google.com14
- 4.kubernetes.io13
- 5.learn.microsoft.com12
- 6.aws.amazon.com9
- 7.cloud.google.com5
- 8.docs.e2enetworks.com5
- 9.builder.aws.com5
- 10.stackoverflow.com4
aurorasre.ai or our GitHub cited on 1 of 50; Aurora named on 0.
Geminigemini-3.5-flash
50 of 50 answered, 185 citations across 108 domains, 2026-10-21.
- 1.medium.com18
- 2.reddit.com9
- 3.stackoverflow.com9
- 4.github.com8
- 5.oneuptime.com8
- 6.netdata.cloud5
- 7.youtube.com4
- 8.amazon.com4
- 9.repost.aws4
- 10.kubernetes.recipes3
aurorasre.ai or our GitHub cited on 1 of 50; Aurora named on 0.
Perplexity leans on github.com, docs.aws.amazon.com, docs.cloud.google.com; Gemini leans on medium.com, reddit.com, stackoverflow.com. The two assistants do not read the same web. Our pages are cited 0 times in 555 citations. Without web search, claude on bedrock, no web search cites us on 0 of 50; what the model already knows is the slow path.
Google's AI Overview on 19 buying-intent queries
A different question: not "how do I fix this" but "which tool should I buy". United States, desktop, 2026-10-21, DataForSEO live SERP, async AI Overview loaded. Google showed an AI Overview on 18 of 19. The sources it cited most, by number of queries: youtube.com (10), incident.io (6), rootly.com (6), github.com (5), reddit.com (5), linkedin.com (4). Our stake: it cites us on 1, and we rank in the organic top 10 on 3.
| Query | AI Overview | Sources it cites |
|---|---|---|
| best open source ai sre | shown | github.comlinkedin.comyoutube.com |
| best ai sre | shown | betterstack.comdatadoghq.comdocs.sadservers.comg2.comgartner.comhyperprobe.coincident.ionofire.ai |
| best ai sre tools | shown | augmentcode.comfluidify.aihackernoon.comhyperprobe.copagerduty.comrootly.comsnowflake.comsystemsdigest.com |
| ai sre tools | shown | augmentcode.comcloudedponderings.medium.comincident.ioresolve.airootly.comsnowflake.comyoutube.com |
| open source ai sre | shown | opensre.inreddit.comunpage.aiyoutube.com |
| ai sre | shown | harness.ioincident.iologz.ioneubird.aireddit.comresolve.airootly.com |
| ai sre agent | shown | augmentcode.comazure.microsoft.comclickhouse.comdatadoghq.comgithub.comincident.ioneubird.aireddit.com |
| open source sre agent | shown | github.comlinkedin.commezmo.comopensre.inreddit.comyoutube.com |
| what are the best ai sre tools | shown | augmentcode.comfluidify.aihyground.aijustaftermidnight247.comsystemsdigest.com |
| top ai sre platforms in 2026 | none | |
| agentic incident response | shown | aimultiple.comakto.ioilert.comlogicmonitor.comowasp.orgpaloaltonetworks.comsweet.securitytechtarget.com |
| aiops open source | shown | digitalocean.comelastic.cofreecodecamp.orggartner.comgithub.commedium.comopenobserve.aiyoutube.com |
| self-hosted ai sre | shown | dev.toedixos.comhackernoon.comlinkedin.comopspilot.comvynelixai.comyoutube.com |
| holmesgpt alternatives | shown | github.comhackernoon.comlinkedin.compodscripts.cowetheflywheel.comyoutube.com |
| resolve ai alternatives | shown | capterra.com.aumedium.comonpage.compeoplemanagingpeople.comrezolve.aisherlocks.aisourceforge.netzentor.ai |
| datadog bits ai alternatives | shown | cbinsights.comenergent.aifluxtail.ioslashdot.org |
| rootly alternatives | shown | incident.iorootly.com |
| best automated post-mortem tools | shown | atlassian.comaurorasre.aidatadoghq.comfirehydrant.comincident.ioreddit.comrootly.comyoutube.com |
| ai root cause analysis tool | shown | adoption.microsoft.combigpanda.iocontentsquare.comcoroot.comeasyrca.comlogz.ioreliability.comtellius.com |
Every debugging question, every source
Domains each assistant cited, in the order it returned them.
| Question | Perplexity | Gemini |
|---|---|---|
Kubernetes My Kubernetes pod is stuck in CrashLoopBackOff. How do I find out why and fix it? | cloud.google.comdocs.aws.amazon.comdocs.cloud.google.comdocs.e2enetworks.comdocs.publishing.service.gov.uk | no sources |
Kubernetes A container keeps getting OOMKilled with exit code 137 in Kubernetes. How do I debug it? | cloudbolt.iodocs.cloud.google.comdocs.dynatrace.comdocs.e2enetworks.comgithub.com | dash0.comfairwinds.comkevinwelter.comkomodor.comkubernetes.recipes |
Kubernetes How do I fix ImagePullBackOff in Kubernetes? | cloud.ibm.comcncf.iodev.todocs.cloud.google.comdocs.e2enetworks.com | kubernetes.recipesmedium.comnashtechglobal.comsignoz.io |
Kubernetes Pod stuck in Pending with FailedScheduling 0/3 nodes are available insufficient cpu. What should I check? | datadoghq.comdocs.cloud.google.comdocs.coreweave.comgithub.comkubernetes.dev | no sources |
Kubernetes Why are my Kubernetes pods getting Evicted and how do I stop it? | arxiv.orgdocs.cloud.google.comgithub.comk8s.infokubernetes.io | no sources |
Kubernetes Kubernetes pod stuck in Terminating state forever, how do I fix it safely? | book.kubebuilder.iocicube.iokubernetes.iolearn.microsoft.comnetdata.cloud | github.comkodekloud.comkubernetes.iomedium.comnetdata.cloud |
Kubernetes What does CreateContainerConfigError mean in Kubernetes and how do I fix it? | devopsness.comdocs.cloud.google.comgithub.comkubernetes.feisky.xyzkubernetes.io | cubeapm.comgroundcover.comkomodor.commedium.comperfectscale.io |
Kubernetes Readiness probe failed: HTTP probe failed with statuscode 503. How do I debug this? | cloud.google.comdocs.aws.amazon.comgithub.comkomodor.comkubernetes.io | no sources |
Kubernetes A Kubernetes node went NotReady. What is the fastest way to find the cause? | cloud.google.comdocs.aws.amazon.comdocs.cloud.google.comdocs.e2enetworks.comkubernetes.io | no sources |
Kubernetes DNS resolution is failing intermittently inside my Kubernetes cluster. How do I troubleshoot CoreDNS? | codewithkarani.comcoredns.iodevscampus.comdocs.byteplus.comdocs.digitalocean.com | no sources |
Kubernetes My Kubernetes Service has no endpoints. Why and how do I fix it? | docs.publishing.service.gov.ukgithub.comkubernetes.ac.cnkubernetes.devkubernetes.io | baeldung.comcontrolplanelabs.comgithub.comlucaberton.comoneuptime.com |
Kubernetes ingress-nginx is returning 502 Bad Gateway. How do I figure out what is wrong? | cloud.ibm.comdevopsness.comdocs.nginx.comgithub.comhost-tracker.com | devopsboys.comkubernetes.recipesnetdata.cloudreddit.comresolve.ai |
Kubernetes Multi-Attach error for volume, volume is already exclusively attached to one node. How do I fix it? | blog.csdn.netcodemia.iodeepwiki.comdocs.cloud.google.comgithub.com | bobcares.combroadcom.comfreshdesk.comgithub.comibm.com |
Kubernetes What is the difference between exit code 137 and 143 in Kubernetes containers? | cast.aicloud.tencent.comdevopsness.comdocs.cloud.google.comdocs.e2enetworks.com | no sources |
Kubernetes My Kubernetes HPA is not scaling and shows unknown for current metrics. How do I fix it? | cloud.google.comdocs.cloud.google.comgithub.comhelp.aliyun.comkubernetes.io | no sources |
Cloud AWS API calls are failing with ThrottlingException Rate exceeded. How do I handle this in production? | aws.amazon.combuilder.aws.comdocs.amazonaws.cndocs.aws.amazon.com | amazon.combobcares.comredbearit.com.aurepost.aws |
Cloud My AWS Application Load Balancer is returning 504 Gateway Timeout. What are the causes and how do I debug it? | aws.amazon.combuilder.aws.comcatalog.us-east-1.prod.workshops.awsdocs.aws.amazon.comdocs.base14.io | amazon.comauthgear.commedium.comreddit.comrepost.aws |
Cloud RDS Postgres says too many connections for role. How do I fix it without downtime? | aws.amazon.comdevopsity.comdocs.aws.amazon.com | atlassian.comdev.tomedium.comnetdata.cloudoneuptime.com |
Cloud EKS worker nodes are not joining the cluster. How do I troubleshoot it? | anywhere.eks.amazonaws.comaws.amazon.comdevfloor9.github.iodocs.aws.amazon.comeksworkshop.com | no sources |
Cloud AWS Lambda function keeps timing out in production. How do I find the cause? | aws.amazon.combuilder.aws.comdocs.aws.amazon.com | no sources |
Cloud DynamoDB ProvisionedThroughputExceededException in production, how do I fix it? | aws.amazon.comdocs.aws.amazon.comrepost.aws | amazon.commedium.comoneuptime.comreintech.iorepost.aws |
Cloud Cloud Run service returning 503 errors intermittently. How do I debug it? | cloud.google.comdocs.cloud.google.com | github.comgoogle.comgoogle.devoneuptime.comreddit.com |
Databases PostgreSQL FATAL: sorry, too many clients already. What should I do right now and long term? | access.crunchydata.comdocs.gitlab.comdocs.selectel.rugithub.comlearn.microsoft.com | no sources |
Databases How do I debug deadlock detected errors in PostgreSQL? | pg.edu.richpgref.devpostgresql.jppostgresql.orgsupport.atlassian.com | no sources |
Databases Postgres replication lag keeps growing on the replica. How do I find the cause? | docs.postgresql.frpostgres.aipostgrespro.compostgresql.jppostgresql.org | no sources |
Databases Postgres warns database must be vacuumed within transactions to avoid wraparound. How urgent is this and how do I fix it? | docs.aws.amazon.comdocs.cloud.google.compostgresql.jppostgresql.org | crunchydata.commedium.commydbops.comnetdata.cloudpercona.com |
Databases Redis OOM command not allowed when used memory > maxmemory. How do I fix it? | help.creoline.comredis.io | no sources |
Databases My Elasticsearch cluster health is red with unassigned shards. How do I recover? | elastic.ac.cnelastic.cogithub.com | datadoghq.comelastic.comedium.comoneuptime.comrepost.aws |
Messaging Kafka consumer lag keeps growing. How do I troubleshoot it? | confluent.iodeveloper.confluent.iodocs.aws.amazon.comdocs.confluent.iodocs.redhat.com | no sources |
Messaging Kafka shows under-replicated partitions. What does it mean and how do I fix it? | docs.aws.amazon.comdocs.cloud.google.comdocs.redhat.comfactorhouse.iogithub.com | no sources |
Messaging RabbitMQ memory alarm is blocking publishers. How do I fix it? | rabbitmq.com | acemq.comdomino.aigithub.comoneuptime.comqarote.io |
Network What is the difference between 502, 503 and 504 errors and how do I debug each one? | datatracker.ietf.orgdevelopers.cloudflare.comiana.orginterserver.netlearn.microsoft.com | no sources |
Network Our TLS certificate expired in production and services are failing. What do I do? | assets-us-01.kc-usercontent.comeff-certbot.readthedocs.ioenterno.iogithub.comnccoe.nist.gov | no sources |
Network Envoy says upstream connect error or disconnect/reset before headers. How do I debug it? | envoyproxy.io | aireiter.comgithub.comhashicorp.comibm.comlast9.io |
Network Cloudflare error 522 connection timed out. How do I fix it? | community.cloudflare.comdevelopers.cloudflare.comelementor.comhost-tracker.comhostinger.com | b12.iocloudflare.comelementor.commedium.compulsetic.com |
Network DNS lookups returning SERVFAIL for our domain. How do I troubleshoot it? | blog.progineous.comcubepath.comdevelopers.google.comdnscale.eudnsdoctor.org | no sources |
Hosts Linux server disk is 100% full in production. How do I safely free space and find the cause? | access.redhat.comcloudstick.iogithub.comhostkey.comlinuxhandbook.com | no sources |
Hosts Too many open files error in production on Linux. How do I fix it? | cs.auckland.ac.nzdocs.kernel.orgdocs.redhat.comhelp.puppet.comkernel.org | no sources |
Hosts Server load average is very high but CPU usage is low. What is going on? | computecentral.incubepath.comdamonsec.comdanielcosenza.comdocs.innozverse.com | no sources |
Runtimes Java app in a container throws java.lang.OutOfMemoryError: Java heap space. How do I debug it? | docs.oracle.comoracle.comoracle.github.io | no sources |
Runtimes Node.js FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory in production. How do I fix it? | beta.docs.nodejs.orgnodejs.org | dev.togitnation.comhalodoc.ioigalia.commedium.com |
Runtimes Our service is timing out waiting for a database connection from the pool. How do I find the cause? | aws.amazon.combuilder.aws.comdoc.netuno.orgdocs.aws.amazon.comnode-postgres.com | no sources |
Delivery Terraform Error acquiring the state lock. How do I fix it safely? | andrewaltimit.github.iobuilder.aws.comdeveloper.hashicorp.comdocs.aws.amazon.comdocs.cloudposse.com | atmos.toolsencore.devenvzero.comhashicorp.comibm.com |
Delivery Helm upgrade fails with another operation (install/upgrade/rollback) is in progress. How do I fix it? | blog.csdn.netblog.helm.shdev.todeveloper.aliyun.comgithub.com | kristhecodingunicorn.commedium.comoneuptime.comstackoverflow.comsuse.com |
Incident patterns How do I stop a retry storm during an outage? | aws.amazon.comazure.microsoft.comd1.awsstatic.comdocs.aws.amazon.comlearn.microsoft.com | google.commedium.comsre.google |
Incident patterns How do I diagnose a cascading failure across microservices during an incident? | aws.amazon.combugsnag.comd1.awsstatic.comdocs.apica.iodocs.aws.amazon.com | arxiv.orgbluepes.comexashard.comgithub.iogoogle.com |
Incident patterns AWS us-east-1 is having an outage. What should my on-call team do right now? | acecloudinterviews.comawsdown.comcloudstatushub.comdocs.aws.amazon.comdocs.lodol.com | amazon.comeon.iomedium.comopen-systems.comwiv.ai |
Incident patterns What is the best way to do root cause analysis for a production incident? | atlassian.comdocs.pagerduty.comhowie-guide.pagerduty.commuc.sre.xyzpagerduty.com | github.comincident.iomedium.compagerduty.comreddit.com |
Tools What are the best AI tools for debugging production incidents? | datadoghq.comdocs.incident.iodocs.rootly.comg2.comincident.io | aurorasre.aigartner.comhoneycomb.iorootly.comsherlocks.ai |
Tools Is there an open source AI SRE agent that can investigate Kubernetes incidents? | dev.codeveloper.hpe.comdevops.gheware.comgithub.comk21academy.com | github.comincidentfox.aikagent.devkubeintellect.commedium.com |
How it is measured
- 50 questions an engineer would type while debugging, written on 2026-09-30 and not edited since. New questions go in a v2 set so runs stay comparable.
- Each assistant gets the question with one system line: you are helping an engineer who is debugging a production system, answer practically and cite your sources. Web search on. Default model for each API.
- A citation is any URL the API returns as a source. "Named" means the answer names Aurora by Arvo AI or links our repo; a bare "Aurora" is not counted, since it usually means Amazon Aurora.
- Rerun weekly. Only the latest run per assistant is shown here; the full history is in the JSON.
Use the data
The whole table is one file, CC BY 4.0: ai-citations.json. Questions, per-assistant domains, and the leaderboard.
If you run an SRE or platform blog and want your domain tracked, or want a question added to v2, tell us.
Related: the agent benchmark and the runbooks we are trying to get cited.
Updated 2026-10-21. Question set v1, frozen 2026-09-30.