A regional HVAC company with eleven locations types its own name into Gemini and asks where to find emergency repair near a specific zip code. Gemini 3.7 Flash does not reach for the directory listing it used to cite. It reads the company’s own location page, pulls the hours, the service area and the phone number, and names the business site as the source. That shift, repeated across thousands of queries, is what we mean by local citation patterns moving toward business-owned websites.
Key Takeaways
- Gemini 3.7 Flash’s agentic capabilities let it navigate a business’s own site directly rather than relying on aggregated directory data, shifting citation share toward owned pages.
- Directories still matter for training-data presence and corroboration, but retrieval presence now depends on whether a brand’s own location pages carry clean, current, structured facts.
- Multi-location brands need a dedicated, schema-marked page per location, not a single directory-dependent location finder, if they want to be the cited source rather than a listed entry.
What Changed in Gemini 3.7 Flash’s Retrieval Behavior
Earlier versions of Gemini leaned heavily on aggregated sources for anything local: directory sites, review platforms, map data. That made sense when the model’s job was to summarize a single page it had already indexed. Gemini 3.7 Flash’s expanded agentic behavior lets it do something different. It can follow a chain of steps within a session: check a brand’s homepage, find the location directory, open a specific city page, and cross-reference the hours listed there against a schema block before answering.
The coding improvements matter here as much as the agentic ones. A model that parses structured data and lightweight scripts more reliably can pull facts straight out of JSON-LD or a rendered table instead of guessing from body text. Gemini still cross-references Search, YouTube and Scholar the way it always has, but the added capability to read a site the way a developer would, rather than the way a crawler used to, means the business’s own page becomes a viable primary source instead of a secondary confirmation.
Why Agentic Crawling Favors Business-Owned Pages Over Directory Listings
Directory listings have a structural weakness that was invisible when models only skimmed snippets: the data is templated and often stale. A location’s hours change, a directory entry does not, and a model that can now verify currency by visiting the source directly has every reason to prefer it. Agentic retrieval rewards the page that is demonstrably current, not the page that ranks highest in a general web index.
This is also where the distinction between Share of Voice and Share of Citation starts to matter for local brands specifically. A business can appear on forty directory sites, driving Share of Voice up, while Gemini cites none of them because none carries verifiable, structured facts the model can act on. Citation share goes to whichever source the model trusts enough to quote, and increasingly that is the business’s own domain.
growth in AI brand presence for NewDay USA, tracked weekly across five engines over six months.See the case study
The Structural Signals That Make a Location Page Citable
Not every owned page benefits from this shift. A location finder built as a single page with a dropdown and JavaScript tabs gives an agentic crawler nothing to point to: there is no stable URL per location, and no clean structured record to quote. Each location needs its own address, a dedicated LocalBusiness schema entry, and hours and service details rendered in plain HTML, not loaded after a user interaction.
Consistency across those fields is what turns a page into a trusted node. Trust, in the way AI models build it, is not a score assigned to a domain. It is a graph of agreement across independent sources, and a location page that matches what appears in the brand’s schema, its sitemap, and any third-party mentions strengthens every one of those connections at once.
Citation share goes to whichever source the model trusts enough to quote, and increasingly that source is the business’s own domain.
Where Directories Still Earn Their Place
None of this makes directories obsolete. AEO has two layers: training-data presence, earned slowly through wide, repeated coverage across the web, and retrieval presence, earned technically through structured, current content a model can fetch live. Directories still do heavy lifting on the first layer. They are part of how a brand accumulates the kind of broad, corroborating mentions that get baked into a model’s general knowledge over time.
What changes is the job directories are asked to do. They stop being the destination a model cites for a specific location fact and become one more node confirming that the brand exists, operates where it claims to, and is described consistently. A multi-location brand still wants its listings accurate everywhere. It just should not expect those listings to be the thing Gemini quotes back to a customer asking for the nearest branch’s hours.
Rebuilding a Multi-Location Site Around Retrieval, Not Just SEO
The practical shift is architectural. Each location gets a permanent URL, not a filtered view of a shared template. Each carries its own LocalBusiness schema with hours, address, phone number and service area, kept in sync with whatever a franchise or operations team updates internally, because a mismatch between the visible page and the schema undercuts the trust graph instead of building it.
Internal linking matters more than it used to. A central locations hub that links to every individual page gives an agentic crawler an obvious path to follow, which mirrors the way these models now navigate sites rather than skim isolated pages. Digital PR and citation-building efforts should point at those specific owned pages rather than at a generic homepage or a directory profile, so the authority a brand earns outside its own site reinforces the exact page it wants cited.
What To Do Next
Start by checking which source Gemini and the other major models currently cite when asked about one of your locations. If it is a directory rather than your own page, that is a retrieval gap, not a content gap, and it will not close by adding more text to the homepage. An AI Visibility Audit will show you, engine by engine, where that citation is currently landing.
From there, prioritize the locations with the highest query volume first. Give each a dedicated URL, add LocalBusiness schema that matches your internal records exactly, and track Citation Analytics over the following weeks to see whether the model’s cited source shifts from the directory to the page you just rebuilt. Citations compound once a model starts citing a page, so the first correct citation on a rebuilt location page tends to make the next one easier to earn.
FAQs
Does Gemini 3.7 Flash Stop Using Directories Entirely?
No. Directories still contribute to the wider training-data presence that helps a model recognize a brand exists and operates consistently. What changes is retrieval presence: when Gemini can verify a fact directly against the business’s own page, it tends to cite that page instead of the directory.
What Is the Difference Between Share of Voice and Share of Citation for a Local Business?
Share of Voice counts every mention of a brand across the web, including directory listings and reviews. Share of Citation counts only how often a model names the brand as the source of an answer. A multi-location brand can have strong Share of Voice through directory coverage while still having near-zero Share of Citation if its own pages are not structured for retrieval.
Do All Locations Need Separate Structured Data, or Can One Schema Cover a Brand?
Each location needs its own LocalBusiness schema entry with its own address, hours and phone number. A single brand-level schema cannot represent location-specific facts accurately, and an agentic crawler verifying a specific address will not find what it needs in a generic block.
How Long Does It Take for a Rebuilt Location Page to Get Cited?
There is no fixed timeline, since it depends on crawl frequency, existing authority and how clean the structured data is. Tracking Citation Analytics on a rebuilt page over several weeks is the most reliable way to see whether the model has started treating it as the source.
