We checked what AI recommends for incident management, and why the uncited tools lose

Structural audit of all 11 tracked tools run October 23, 2026 via direct HTTP checks. Presence numbers come from DevTune's public tables; the 90-day report and the 30-day live page use different windows and are never mixed.

AI answers point buyers toward a short list of incident management tools. incident.io, Rootly, and PagerDuty take most of the spots. When we checked which citable assets the eleven tracked tools in this category actually maintain, the pattern was uncomfortable: the tools buyers never hear about are not all missing the basics, and having the basics does not get you cited. The tools with the most complete machine-readable setup sit at or near zero.

That is the short answer. The rest of this note shows the data behind it: what AI currently recommends for incident management and on-call, what each of the eleven tools maintains in terms of machine-readable assets and comparison pages, and what the difference means if your tool is the one getting left out.

This is a diagnosis, not a leaderboard. Other publications already rank these tools. We are asking a different question: why do the uncited tools lose?

What AI recommends for incident management today

ChatGPT Search, Perplexity, Gemini Search, Google AI Mode, and Bing Copilot mostly point buyers to three brands: incident.io, Rootly, and PagerDuty. DevTune's public State of AI Search report measured 1,255 answers for this category between April 2 and June 30, 2026, and found incident.io present in 47.9% of them, Rootly in 32.7%, and PagerDuty in 26.8%. The category median across all tracked brands was 12.4%.

DevTune runs a live public page for this category with a rolling 30-day window, and the same three names still lead it as of late October 2026: incident.io at 42.7% presence, Rootly at 36.7%, PagerDuty at 34.0%. Two details from that page:

• The rest of the field has nearly disappeared. In the 30-day window, ilert sits at 2.0% presence, OneUptime at 0.7%, and both Spike.sh and Squadcast at 0%.
• Being cited is not the same as being named. DevTune distinguishes presence (your pages get cited as sources) from brand mentions (the answer names you as an option). PagerDuty was present in 34% of answers but mentioned in 0%. incident.io was present in 42.7% and mentioned in 66%.

The full 30-day picture, from DevTune's public brand table:

Rank four in that table is Opsgenie, a product Atlassian stopped selling on June 4, 2025 and will end support for on April 5, 2027, per Atlassian's own migration page. In the 30-day window, its pages still get cited in about one answer in seven. AI answers run on the evidence that existed when their sources were written, so products with large documentation footprints keep getting cited long after the market has moved on.

(All DevTune numbers in this section come from the two public pages linked at the end, with their respective windows. The 90-day report and the 30-day live page use different sample sizes, so treat small gaps between them as measurement noise.)

What we checked, and how

We tested every tracked brand in the category for the citable assets an answer engine can actually use. Eleven tools, six checks each, all run with direct HTTP requests on October 23, 2026:

1. llms.txt at the domain root, the emerging index file for machine readers
2. llms-full.txt, the single-file variant
3. robots.txt rules for AI crawlers (GPTBot, ClaudeBot, PerplexityBot, and friends)
4. Competitor comparison or alternatives pages in the sitemap
5. Public benchmark or original-research content
6. Documentation portal reachability (HTTP status and page title)

These are structural facts, not model sampling. Every check either returned a status code and content we could inspect, or it did not. Where a server returned an HTML error page with a 200 status, we counted that as missing.

What the audit found

Seven of the eleven tools publish a valid llms.txt. Four publish an llms-full.txt. Two explicitly welcome AI crawlers in robots.txt by name; most of the rest allow them by default. One tool's documentation sits behind a login wall.

PagerDuty, Opsgenie, FireHydrant, and Grafana publish no comparison pages at all. Squadcast's sitemap could not be audited because it redirects to SolarWinds' site-wide sitemap.

Cross-reference this table against the presence numbers and two findings stand out.

Finding one: the missing basics explain the very bottom, not the middle. Squadcast is the clearest case. Its domain serves no valid robots.txt and no llms.txt, its sitemap redirects to SolarWinds' site-wide sitemap, and its docs domain redirects to a Google login page. Nothing about that setup invites a machine reader. It sits at 0% presence. That part of the diagnosis is simple: if a crawler cannot reach your docs without a login, you are not in the game.

Finding two: the assets alone do not produce citations. Spike.sh maintains an llms.txt, an llms-full.txt, two comparison pages, and a fully open robots.txt. In the 30-day window its presence is 0%. OneUptime goes further than anyone: it labels its robots.txt "AI / LLM crawlers are welcome", publishes both llms files plus a separate llms.txt for its docs, and maintains thirty comparison pages. Its presence is 0.7%. The checklist behaves like a prerequisite; the citations come from somewhere else.

What does the citation data say the winners actually feed on? In the 30-day window, DevTune recorded the content mix behind this category's 2,487 citations as: blog posts 31.6%, documentation 24.8%, comparison pages 19.1%, best-of lists 10.0%, product pages 8.1%. Company-owned and commercial sources account for 85.5% of what gets cited, editorial coverage 9.1%, community threads 1.3%. And in the 90-day report, DevTune's own attribution for incident.io's lead reads: "Category education drives most of the leader's citations." Nearly 88% of the citations the leader receives point to blog posts, not docs or product pages.

That is the real gap. The tools that lose are not missing a config file; they are missing the surrounding body of evidence that answers draw on when a buyer asks a question: the category explainers, the worked comparisons, the third-party lists, the benchmark posts with methodology. incident.io wins with a blog. Rootly wins with comparison pages that Gemini Search cites. PagerDuty wins on documentation alone, which is why it gets cited as a source without ever being named as an option. There is more than one path in, and all of them run through published, findable evidence.

Why the uncited tools lose

Pulling the two data sets together, the losing pattern in this category looks like this:

• The tools at zero maintain, at best, machine-readable versions of a thin site. When an answer engine assembles "best incident management tools," it draws on category education and comparison material. If you have published none that ranks, you are not in the pool of candidates.
• Walls sit where doors should be: login-walled docs (Squadcast), missing robots.txt files, and domains folded into a parent brand (Opsgenie into Atlassian, Squadcast into SolarWinds) all cut the path between your content and the answer.
• Machines can read Spike.sh and OneUptime, but they have no reason to cite them. Without the surrounding proof, the llms files index an argument nobody has made yet.
• Stale options persist. Opsgenie's fourth place, eighteen months after its end-of-sale date, shows that answers do not refresh on their own. That cuts both ways: whatever position your tool holds in AI answers today is the one it will keep until someone publishes better evidence.

One caveat we want on the record: this audit measured structure, not coverage. We did not count third-party mentions, review sites, or community threads, and DevTune's data shows those play a role. Our structural findings describe what is missing on the tools' own domains. They cannot prove which lever moves a specific brand's numbers, and we would not trust anyone who claims a fixed checklist guarantees citations.

What to do if your tool is the one getting left out

The sequence that matches the data for this category:

1. Open the doors first. Valid robots.txt, an llms.txt that lists your key pages, and documentation a crawler can reach without a login. Most of the eleven already clear this bar; if yours does not, nothing else on this list matters much.
2. Build the page the citation mix is hungry for. In this category that means category education (what incident response is, how to choose) and honest comparison pages. The leaders' citation profiles are dominated by exactly these formats.
3. Show up where third-party lists are built. One in ten citations in this category points at best-of lists and review pages. If those pages only know the same three brands, they will keep recommending the same three brands.
4. Keep it fresh. Opsgenie's continued presence shows answers run on stale evidence for a long time. Publish dates, current pricing, and maintained comparisons give an answer engine a reason to update its story.

If you want this same check run against your own tool, that is what CiteGap's free audit does: it inspects the citable assets on your domain, compares them against what actually gets cited in your category, and sends you the gaps in plain language.

Run the free lite audit

Frequently asked questions

Which incident management tools does AI recommend most?

Across DevTune's 90-day report window and its current 30-day live table, the same three lead: incident.io, Rootly, and PagerDuty. incident.io leads both windows; PagerDuty is cited as a source more often than it is named as a recommendation.

Why is my incident management tool never mentioned by AI?

In this category, the most common structural causes are a missing evidence web (no category explainers or comparison pages), technical barriers (login-walled docs, no llms.txt, restrictive robots.txt), or being a newer brand in a field where AI answers lean on established sources. Our audit of all eleven tracked tools found 7 of 11 with valid llms.txt and 6 of 11 with comparison pages; the tools at zero presence were missing either the structure, the evidence, or both.

Does publishing llms.txt guarantee AI citations?

No. Spike.sh and OneUptime both maintain strong llms assets and sit at or near zero presence. llms.txt and open robots.txt look like entry requirements. What the citation data rewards in this category is category education and comparison content that answers can quote.

How can I check my own devtool's citability?

You can replicate the structural checks yourself (llms.txt, robots.txt, docs reachability, comparison pages in your sitemap), or run CiteGap's free audit, which does the checks and adds a category-level comparison. For brand-level presence numbers, DevTune publishes the live table for this category that this article draws on.

Sources

• DevTune, State of AI Search for Dev Tools (data window April 2 to June 30, 2026; 43 categories, 1,255 measured answers for this category): devtune.ai/dev-tools-state-of-ai-search
• DevTune, Incident Management & On-Call live brand table (30-day rolling window, accessed October 23, 2026): devtune.ai/verticals/incident-management-and-on-call
• Atlassian, Opsgenie migration page (end of sale June 4, 2025; end of support April 5, 2027): atlassian.com/software/opsgenie/migration
• Structural audit of 11 incident management tools: direct HTTP checks run by CiteGap on October 23, 2026 (llms.txt, llms-full.txt, robots.txt, sitemaps, docs reachability)