GPT-5.6 Sol, Terra, and Luna: What's Actually Known About Citation Rates

GPT-5.6 shipped in mid-July 2026 with a three-tier model structure: Sol, Terra, and Luna. Within days, AEO content was circulating specific citation rate differences between Sol and Terra, framing them as evidence that businesses need to urgently optimize for the tier their customers land on.

Some of those numbers are wrong. Not wrong in direction -- there probably is a meaningful Sol/Terra citation gap. But the specific benchmarks appearing in AEO practitioner content as of July 2026 are measuring something entirely different than what their authors think.

The Three Tiers and Who Gets Each One

The access mapping for GPT-5.6 is confirmed across multiple independent sources:

| User type | Model | |---|---| | Free / Go | Terra (Luna in API high-volume) | | Plus | Sol at medium effort | | Pro / Business / Enterprise | Sol at medium or high effort |

This structure matters for any business trying to understand its AI visibility. Whether ChatGPT mentions your business depends heavily on the subscription tier of the person asking. Sol produces longer sub-query fan-out -- more searches per prompt, a larger candidate pool of sources, a higher probability of surfacing any specific business. Terra runs a more constrained retrieval path.

There is also a silent rerouting layer. Under GPT-5.5, 16% of free-tier prompts were moved to the paid-tier model without the user knowing. With three tiers in GPT-5.6, silent rerouting between Luna, Terra, and Sol adds another source of unpredictability in what any specific user sees for the same query.

The Numbers in AEO Content Are From the Wrong Study Type

In our 2026-07-17 investigation of the Sol/Terra citation benchmarks circulating in AEO practitioner content -- blogs, developer resources, and AEO tool marketing pages -- we found a specific problem with how these numbers are being sourced.

The benchmarks being cited as evidence of "Sol vs. Terra web citation behavior" come from in-context long-document retrieval studies. These studies test something different. A developer provides a long document inside the model's context window. The model is then asked to retrieve specific information from that document. "Citation accuracy" in this context means correctly attributing retrieved content to the right source document within the prompt.

That tests whether Sol can track provenance within a conversation. It does not test how often Sol links to your business's Yelp listing when a user asks ChatGPT "best plumber in Austin."

A result like "Sol achieves 97% routing accuracy in primacy-positioned document retrieval" tells you nothing about how frequently Sol chooses to surface a local business in a recommendation query. The two measurement contexts share vocabulary -- "citation," "source selection" -- but they describe completely different behaviors.

What IS Actually Confirmed

The GPT-5.5 Thinking vs. Instant gap is measured and real, and it is the best available baseline for understanding what GPT-5.6's tier structure likely produces.

Writesonic ran 50 prompts across 16 categories, analyzed 11,469 web search results and 1,257 citations from the June 2026 GPT-5.5 deployment. Their findings:

- Paid tier (Thinking): 47% brand site citation rate - Free tier (Instant): 6% brand site citation rate

That is a roughly 40-percentage-point gap between tiers for the same prompts. SISTRIX confirmed the scale of the shift from a different angle: 47% of citations redistributed within 48 hours when the free-tier default model switched.

As of July 17, 2026 -- when our methodology rec on GPT-5.6 audit scope was filed -- the equivalent studies measuring Sol vs. Terra web citation rates have not been published. Writesonic opened their GPT-5.6 measurement window on July 16. SISTRIX opens July 21. Specific Sol/Terra citation percentages for web search behavior do not yet exist as confirmed data.

The directional expectation is that Sol will cite more than Terra. The mechanism is consistent with the GPT-5.5 pattern: Sol generates more sub-queries per prompt, reaches more sources, and has more opportunities to surface any given business. The architectural analogy strongly suggests the gap will be significant. But "significant" is not the same as the specific numbers that have been cited in AEO content without a verifiable methodology or study reference.

Why This Creates a Real Problem for Businesses

If you're making AI visibility decisions based on AEO content that cites specific Sol/Terra citation rates, you may be acting on numbers that don't describe the behavior you think they describe.

The correct framing as of today:

- There is a real tier gap in ChatGPT. Paid-tier users see meaningfully more business citations than free-tier users. This is documented at the GPT-5.5 level. - The GPT-5.5 Thinking/Instant baseline -- 47% paid vs. 6% free -- is the best available proxy for what the Sol/Terra gap will look like. - The GPT-5.6 Sol/Terra gap will be confirmed in the coming weeks when the Writesonic and SISTRIX studies close. - Reasoning mode adds a third layer on top of tier structure. Semrush's June 30, 2026 study (100 prompts, 20 buyer journeys) found that 74.4% of sources differ between minimal and high-reasoning mode for the same query. A Plus subscriber using extended thinking is drawing from a materially different citation pool than a Plus subscriber in default mode -- even though both are on Sol.

This means "is my business visible in ChatGPT" is actually three nested questions: which tier is the user on, which reasoning mode did they use, and which specific query formulation did they run.

What to Do While the GPT-5.6 Data Comes In

The tier structure and the confirmed GPT-5.5 gap don't change the tactical reality for most local businesses, even without precise Sol/Terra percentages.

Under GPT-5.5 Instant -- and likely under Terra -- brand-owned website citations dropped to near-zero for most local service categories. Third-party presence became the primary lever: directories, review platforms, earned media that appears in organic search results for the query. That conclusion doesn't require Sol/Terra citation percentages to act on. Whether the Terra brand citation rate turns out to be 4% or 10%, the fix is the same: directory presence, NAP consistency across platforms, review volume above the competitive threshold for your market, and category-specific listing optimization.

What WILL change when the Sol/Terra data is confirmed: whether it's worth running separate audit queries at the Sol and Terra tiers to show clients the gap. Our current methodology (per the July 16 rec) specifies which tier was queried, but runs only the Sol/Plus tier as the default. If the Sol/Terra gap exceeds 20 percentage points -- consistent with the GPT-5.5 pattern -- a dual-tier audit output will likely become standard.

If you want to see where your business stands right now across platforms and query types, a Signal Check will show your current citation status. That gives you a concrete baseline before the GPT-5.6 data settles.