Gemini Cites the Same Business 7% of the Time on Repeated Queries

If you've run an AI visibility audit and noticed your Gemini score feels unreliable -- you appeared last week and not this week for the same query -- you're not misreading the data. Our September 23, 2026 analysis of the Steady Demand AI Citation Ledger (14,472 local citations, July 2026 data) documented exactly why.

The same top business appears in only **7% of repeated identical Gemini queries**. On the Google local pack -- the traditional local search surface that Gemini nominally supplements -- that figure is 90%.

Gemini is not a noisy version of Google search. It's a fundamentally different citation environment.

What a 7% consistency rate actually means

A business visible in Google's local pack for "dentist near me" shows up roughly 9 times out of 10 when that query runs again. The ranking infrastructure is deterministic enough that you can count on it.

Gemini's organic citation layer doesn't work that way. If a business is in the citation pool for a given query, it will appear some fraction of the time -- but the fraction is low. For the top businesses in the Steady Demand dataset, that fraction averaged 7% across repeated identical queries.

This isn't a measurement artifact or a non-determinism problem specific to how we run tests. Our July 2026 methodology rec (filed 2026-07-30), analyzing the SISTRIX 17-week citation study (82,619 prompts, 1,548,213 snapshots), documented that Google AI Mode rotates 56% of its source pool each week. The Steady Demand data puts this in terms of individual business presence: even a business that Gemini is clearly willing to cite will be absent from most query runs.

The practical consequence: Gemini audits measure a different thing than ChatGPT or Perplexity audits. A Gemini citation result is a sample from a low-probability distribution. "Not appearing in this run" doesn't mean you're not in Gemini's citation pool. It means you're in a pool with low per-query hit rate.

Why Gemini's source architecture differs from ChatGPT

The consistency problem becomes clearer when you look at how Gemini selects sources in the first place.

The Steady Demand citation ledger (14,472 citations, July 2026, filed in our September 23, 2026 research update) shows Gemini's local citation source split:

- Business website: **59.9%** - Reddit and community platforms: 13.7% - General directories: 10.4% - Local-service directories (Angi, Thumbtack): 10.3%

The business's own website accounts for nearly 60% of Gemini's local citations. For every citation pulled from local-service directories like Angi or Thumbtack, Gemini pulls approximately six from the business's own site.

ChatGPT's architecture is inverted. Third-party directories dominate because ChatGPT has structural data feed partnerships -- Foursquare POI data (since December 2024), Yelp's 330M reviews and 8M listings (since July 2026). A business absent from those directories is structurally absent from ChatGPT local responses regardless of how good its website is.

The Steady Demand data also quantified the divergence between platforms directly: **ChatGPT and Gemini share only 8% of citations on identical queries**. Ask the same local category question to both platforms and expect completely different citation sets 92% of the time.

That 8% overlap figure should change how you think about multi-platform AI visibility. The implicit assumption in most AEO strategies is that there's a shared citation pool and the goal is to appear in it. The actual picture is two nearly independent citation ecosystems, each with different source weights, different consistency characteristics, and different fix levers.

What to do differently for Gemini

The 59.9% website figure and 7% consistency finding point to the same strategy: website signals are both the primary lever and the most direct way to move the consistency number.

A business with authoritative on-site structure -- LocalBusiness schema with correct @id, sameAs links to directory listings, service pages with direct-answer content structure -- is a stronger candidate for Gemini's citation pool than a business with a thin website and a full directory presence. The directories still matter for entity consistency signals that propagate across all platforms, but they're not where Gemini's citations are coming from.

The framing shift this requires: optimizing for Gemini means treating your website as the primary citation surface, not a supplementary signal. Schema markup, well-organized service pages, entity clarity -- these are not Phase 2 website polish. They are the primary fix for a low Gemini score.

A few things that don't change the Gemini analysis:

**More directories won't fix a low Gemini score.** If your Gemini citations are weak and your ChatGPT citations are weak, you might have overlapping gaps in directories and website signals both. But adding Yelp and Foursquare to fix Gemini is a mistaken priority. Fix the website signals first.

**The 7% consistency rate doesn't mean Gemini optimization is futile.** It means the optimization goal is to appear in a higher share of that 7% -- to build enough on-site authority that Gemini surfaces you whenever its rotation passes through your category. A business in the citation pool with strong on-site signals will appear more consistently than a business at the margin.

**For home services, the Connected Apps layer is separate from this.** Our August 2026 methodology rec (filed 2026-08-17) documented that Angi and Thumbtack became Gemini Connected Apps on August 12, 2026. This is an action-routing layer -- Gemini routes homeowners to Angi and Thumbtack contractors when they describe home projects. It bypasses the organic citation consistency problem. Being on Angi and Thumbtack does not fix organic Gemini citation frequency for non-home-services businesses. For home services specifically, the Connected Apps presence is a distinct, stable signal layer on top of (not a replacement for) organic citation optimization.

What this means for your audit results

A Sourcepull audit scores Gemini separately from ChatGPT and Perplexity because the signals that move each score are different. A business with a strong directory presence and weak website content will often score higher on ChatGPT than Gemini. A business with a well-structured site and thin directory presence will often score the opposite. The per-platform breakdown is where the fix priority emerges.

If you've run a Signal Check and your Gemini score is low while your ChatGPT score is reasonable, the gap is almost certainly on-site: schema, entity markup, content structure. That's where to start -- not another directory listing.

If you're seeing your Gemini presence fluctuate run-to-run, the 7% consistency data says that's the platform, not a measurement problem. The goal isn't to fix the inconsistency. It's to be consistent enough in Gemini's source pool that the 7% of runs where you appear is yours.