Google AI Overviews for B2B Software: How to Rank in AI Overviews and AI Mode

Google AI Overviews are the one engine in the AI Shortlist Index protocol with no API, so there is nothing to call. The protocol captures a US desktop results page for the prompt text through a third-party provider. It reads the AI Overview off the page if one is there, and records that it was absent if not.

That last part matters most, because for Google whether an answer appears at all is a measurement, and absence is recorded rather than zeroed. This guide covers the difference between AI Overviews and AI Mode as Google describes them, and what Google says about eligibility and query fan-out. It ends with what that implies for a property-ops software vendor.

The commercial version is how to appear in Google AI Overviews for software searches. The mechanism shared by all the engines is on the AI search optimization pillar.

What is the difference between AI Overviews and AI Mode?

Google now shows generated answers in two places. Google’s Search Central documentation on AI features describes AI Overviews as a way for a user to get the gist of a complicated topic quickly, with links for deeper exploration. It describes AI Mode as the experience for queries where further exploration, reasoning or complex comparisons are needed.

An AI Overview sits at the top of an ordinary results page. AI Mode is a separate tab and a conversation.

The Index protocol measures AI Overviews and not AI Mode. AI Mode is a logged-in, conversational surface with no stable page to capture. The Index’s rule is that every engine needs a capture a reader can check: a documented API, a licensed data provider, or a stored capture of the app itself.

When a repeatable capture path for AI Mode exists, adding it will be logged in the Index changelog.

Both features draw on the same index. Google’s documentation says the links shown in AI Overviews and AI Mode come from pages that are indexed and eligible to appear in Search with a snippet. There are no additional requirements, so what applies to Overview eligibility applies to AI Mode too.

What makes a page eligible?

Google is unusually direct about what is required: its AI features documentation says there is no special schema.org structured data to add. No new machine-readable files, AI text files or markup are needed to appear in AI Overviews or AI Mode. A page needs only to be indexed and eligible to be shown in Search with a snippet.

A vendor being sold “AI Overview optimization” as a separate product line should read that sentence twice.

The controls run in the other direction. Google documents that the nosnippet, data-nosnippet and max-snippet directives, and noindex, limit what content can appear in its AI features, the same way they limit snippets in ordinary results.

A vendor whose pricing table sits inside a data-nosnippet block, or whose integration list is on a page with a tight max-snippet, has opted its most quotable facts out. Check for these before checking anything else.

Google’s documentation also separates Google-Extended, the robots.txt token that controls AI training and grounding in some of Google’s other systems, from AI Overviews in Search, which Google-Extended does not affect. Blocking Google-Extended does not remove a vendor from Overviews, and allowing it does not add one. The crawler that matters is Googlebot; the AI crawlers guide has the full picture.

How does query fan-out work?

Google’s documentation says both AI Overviews and AI Mode may use a technique it calls query fan-out, issuing multiple related searches across subtopics to develop a response. That lets them display a wider and more diverse set of links than a traditional results page for the same query.

This is the same behaviour the ChatGPT guide documents from the other side, where a single maintenance prompt turned into five search calls and 124 consulted URLs.

For a B2B software query the fan-out is easy to picture. A PM’s question about maintenance software for 300 doors becomes searches for the category, the door-count fit, integrations with whichever PMS is common at that size, and pricing.

We expect the pages cited in an Overview for a property-ops prompt to be drawn from those sub-queries rather than from the literal question. That would mean a vendor’s integration page can be cited for a prompt that never mentions integrations; the first Google edition will show whether the cited links bear that out.

The implication runs the other way too. A vendor cannot rank for the 17-word question, and does not need to.

It needs to be the clearest indexed answer to the two or three sub-questions the fan-out produces, which are ordinary search queries with ordinary results pages. The Index prompt library exists partly so vendors can work out what they are for their own category.

What does a PM’s prompt look like as a Google query?

A prompt written the way a property manager asks, with a door count, a property type and a PMS in it, is a long conversational query. “What’s the best maintenance coordination software for a property management company with around 300 doors?” goes into the search box as a 17-word question. That is what a PM who asks questions instead of typing keywords would do.

Our hypothesis is that these queries trigger an AI Overview more often than short category keywords do, because they read as questions needing synthesis rather than navigation. We expect the Overview rate to vary sharply across the six prompt types in the library, with problem-led prompts producing Overviews most often and comparison prompts least.

The protocol records whether an Overview appeared per prompt per run, so the rate by prompt type can be computed rather than guessed.

What the ChatGPT runs suggest

The one run whose consulted list is visible is a ChatGPT answer to the 300-door maintenance prompt captured through the OpenAI API on 3 September 2026. It is retired from the edition and kept in the archive, and it named Property Meld first and Latchel second and drew most heavily on reddit.com, appfolio.com and buildium.com.

The 25 app captures the September edition rests on show only the links in each answer. In the maintenance category those went mostly to vendor sites, led by latchel.com, propertymeld.com and lula.life. Whether a Google Overview for the same query cites the same domains is an open question.

We expect the overlap to be partial. The PMS vendors’ content and the product sites are indexed and snippet-eligible. Much of the forum material is either thin or marked up in ways that make it a poor snippet source.

What should a property-ops vendor do?

In order of influence.

1. Fix eligibility first. Every page that states a fact a PM would ask about must be indexed, return a 200, and carry no snippet restriction on the facts themselves. This is a ten-minute check in Search Console and a crawl. It is where “we never show up in Google AI” can turn out to be a nosnippet directive left in an old template.
2. Answer the sub-questions on their own pages. Integrations by PMS name, on a page whose title says so. Pricing and portfolio fit, on a page whose first paragraph says so. What the product does that the PMS’s built-in work orders do not, in plain text. The fan-out will look for each of those separately.
3. Get on the pages the sub-questions already return. The PMS marketplaces, the trade press, the association directories. The category’s source map will show which of those an Overview cites once Google runs exist.
4. Measure both numbers. The Overview rate for your category’s prompts, then your mention rate within the Overviews that appeared. The AI visibility guide defines the metrics. Search Console will not separate them for you. Google’s documentation says traffic from AI features is reported in the Performance report under the Web search type, mixed into overall search results.

What will the Index record from Google?

Nothing yet. The September 2026 edition’s numbers come from ChatGPT, and the Google engine has no runs, so it is excluded from every denominator rather than scored as zero. The capture path is written and disclosed on the Index methodology page.

When it runs, the protocol captures a US, non-personalized, desktop Google results page for the exact prompt text through a third-party provider. It extracts the AI Overview text and the links in it, and records whether an Overview appeared at all. The organic top results and the cost are stored with each run.

Absence is recorded, not zeroed

This is the design decision that makes the Google engine different from the others. When ChatGPT is asked a prompt, an answer comes back and a vendor is either in it or not. When Google is asked a prompt, there may be no AI Overview on the page, and the reason is not the vendor’s.

The Index methodology states that when Google AI Overviews do not appear for a query, absence is recorded as “no AI Overview”, not as a zero. In practice the presence flag is false for that run and the run does not count in the vendor’s Google mention rate denominator. The share of prompts that produced an Overview can be computed as its own number.

A vendor with a low Google mention rate and a category with a low Overview rate are different findings. Folding one into the other would make the Index lie in both directions. The same rule extends to degraded engines: failed calls are stored as failure records with a reason, and nothing becomes a silent zero.

Limits, and what changes month to month

Third-party capture. The protocol reads Overviews off a captured results page, not from a Google API, and the capture depends on a provider Google does not endorse. The provider, the location settings and the device type will be recorded per run, and any change is a logged capture-method change.

No fallback. When the provider fails, the engine ships degraded for the edition. It does not get retried through a different path, because a different path would not be comparable.

AI Mode is not captured. A vendor’s Google row, once it exists, will describe AI Overviews on the standard desktop results page and nothing else.

Results pages move, because whether an Overview appears for a given query can change from day to day, and Google tests layouts continuously. Three runs on separate days and the presence flag make that visible per prompt. The Overview rate will move more month to month than any vendor’s mention rate.

Engines are US-only and unpersonalized in the Index. A logged-in PM in Texas with a search history will see a different page. The Index measures the baseline.

Related guides

The Gemini guide covers the other Google engine, which does have an API and returns grounding chunks that resolve to final URLs. The ChatGPT guide has the archived API run to read, whose trail of consulted sources most resembles a fan-out in the open. The Perplexity guide covers dated search results and citation ordering.

The full protocol, including the seed-vendor exclusion and the accuracy gate on extraction, is on how we measure AI recommendations, and this site’s own scorecard is at the reference implementation.

Changelog

• 2026-10-28: Reformatted with a short answer, question headings, the vendor steps as a numbered list, and an FAQ. Retitled toward “how to rank in AI Overviews”. The capture section now says Google has no runs yet and describes the protocol in the future tense.
• 2026-09-26: The September figures now cite the 25 ChatGPT app captures per category; the API run of 3 September 2026 is described as retired from the edition and archived.
• 2026-09-19: Corrected the capture rule stated in the AI Mode section. The Index no longer queries only APIs and licensed providers; ChatGPT is captured from its app, and what AI Mode lacks is any checkable capture at all.
• 2026-09-03: Rebuilt for the AI Shortlist Index scope. Added AI Mode and the distinction Google draws, the third-party SERP capture, the presence flag and the absence-not-zero rule. Added query fan-out and the eligibility and snippet controls from Google’s AI features documentation.
• 2026-06-10: First published as a general Google AI Overviews guide.