ChatGPT Brand Citations Are Recovering — but Only for Paying Users

In May 2026, GPT-5.5 Instant became ChatGPT's default model for free users, and brand site citations on that tier collapsed. Writesonic's pre/post analysis documented the drop: brand website citations fell from 13% to 6%, and the site: operator that ChatGPT had used to proactively seek out brand domains dropped from 40.5% to 12.6%. The practical guidance that followed was clear -- for the free-tier majority, third-party directories and review platforms had displaced brand-owned content as the primary ChatGPT citation lever.

That analysis was correct. It still is, for most of your potential customers. But something shifted in July 2026 that is already generating confusion in AEO practitioner circles -- and the confusion matters because it could push businesses toward the wrong fix plan.

What GPT-5.6 Sol Changed

Writesonic published a GPT-5.6 Sol citation study in mid-July 2026. In our July 20, 2026 investigation of the finding (Scout session 82, `research-vault/methodology-recs/2026-07-20-gpt56-sol-brand-site-citation-reversal.md`), we traced the headline result: GPT-5.6 Sol's site: operator usage is back at approximately 71% -- nearly twice GPT-5.4's rate and 5.6 times GPT-5.5's rate.

| Model | Site: operator usage | |-------|---------------------| | GPT-5.4 | ~40.5% | | GPT-5.5 Instant | ~12.6% | | GPT-5.6 Sol | ~71% (directional, July 2026) |

The site: operator is the mechanism that mattered. When ChatGPT uses it in a web search sub-query, it is specifically seeking out a brand's own domain. When it dropped under GPT-5.5, brand-owned pages stopped entering the candidate pool unless they happened to rank organically for the query. With Sol, that proactive brand-seeking behavior is back -- at a rate higher than before GPT-5.5 arrived.

This directionally means brand site citations are up in Sol-tier responses. Writesonic's full study remains inaccessible from our research environment as of July 23, so the exact brand citation percentage has not been confirmed -- the 71% figure is from search snippets, not the primary article. The direction is not in question. The precise number is.

Who Sol Actually Reaches

This is where the July 2026 picture diverges sharply from how it is being discussed.

In our July 23, 2026 follow-up investigation (Scout session 85, `research-vault/methodology-recs/2026-07-23-gpt56-sol-free-tier-clarification.md`), we documented what model each ChatGPT user segment actually receives in standard plain-chat conversations:

| Plan | Model in plain chat | |------|---------------------| | Free / Go | GPT-5.5 Instant | | Plus | GPT-5.6 Sol | | Pro / Business / Enterprise | Sol (higher tiers selectable) |

The free tier does not get Sol, or any GPT-5.6 variant, in standard conversations. It gets GPT-5.5 Instant -- the same model from May 2026 that drove the brand citation collapse. Sol is only available in plain chat to Plus subscribers and above.

This matters enormously for what the Sol reversal actually means in practice. The majority of ChatGPT users are on the free tier. Most consumer queries -- a homeowner searching for a plumber, a patient looking for a dentist -- come from free accounts. Those queries are still answered by GPT-5.5 Instant, where site: operator usage is at 12.6% and brand site citations are near zero outside a handful of categories.

Our July 23 session concluded explicitly: the third-party-first framing from our June 16 methodology rec remains accurate for the majority of real consumer queries. Sol is real. Sol is consequential. Sol is just not what most of your potential customers are using.

A Mischaracterization Spreading Through AEO Content

Our July 17, 2026 edge case investigation (Scout session 80, `research-vault/edge-cases/sol-terra-in-context-vs-web-citation-conflation-2026-07-17.md`) flagged a specific mischaracterization circulating in developer and AEO practitioner content in mid-2026 -- and it is directly relevant to how the Sol reversal is being received.

You may encounter numbers like "Sol achieves 97% citation accuracy" or "Sol dramatically outperforms Terra on citation rates." Those figures are real. What they measure is not web citation behavior.

They come from in-context long-document retrieval benchmarks: tests where a developer loads a document into the model's context window, then asks the model to attribute retrieved content to the correct source inside that prompt. Citation accuracy in this context means correctly identifying which supplied document a piece of information came from. It has nothing to do with how often GPT-5.6 Sol cites external websites when answering a user's open-ended question.

For AEO purposes, the relevant question is web citation behavior: when a user asks ChatGPT "best electrician in Denver," how often does Sol cite a brand website versus a directory listing versus a review platform? That data from a controlled study did not exist as of July 17, 2026 -- the Writesonic and SISTRIX measurement windows had just opened. Practitioners citing "Sol 97% citation accuracy" as evidence that Sol-tier users will find brand content are drawing on the wrong benchmark entirely.

The conflation is generating both over-investment in brand site work (for businesses whose buyers are mostly free-tier consumers) and under-investment (for B2B buyers whose ChatGPT usage skews Plus and above).

What This Means for Fix Plan Sequencing

The Sol finding does not reverse the directory-first recommendation for most SMBs. It adds a more nuanced second layer for specific segments.

For businesses whose customers are primarily consumer-intent free-tier ChatGPT users -- which describes most local service categories -- GPT-5.5 Instant is answering their queries. Third-party directory presence remains the primary ChatGPT lever. Schema and brand site optimization still produces its strongest citation returns on Gemini and Perplexity, not on the ChatGPT free surface.

For businesses whose target buyers are more likely to be on paid ChatGPT plans -- professional services, B2B, high-income homeowners who subscribe to Plus for work research -- the Sol reversal is relevant. Brand site schema, well-structured service pages, and entity signals that Sol's broader sub-query fan-out can retrieve become defensible Phase 2 investments again.

The conclusion from our July 23 investigation: do not reprioritize brand site schema over directory presence for general SMB fix plans based on the Sol finding. If the client's buyer profile is demonstrably Plus-skewing, brand site work moves up with stronger justification. Otherwise, the sequencing holds -- directory presence first, brand site work second, with the return on each shaped by who is actually using ChatGPT to find you.

A Sourcepull Signal Check breaks out scores per platform -- ChatGPT, Perplexity, Gemini, and Claude -- so you can see whether a ChatGPT gap is a directory problem or a brand site problem, rather than applying the Sol finding uniformly to a split-tier user base. If your ChatGPT score is low and your Gemini score is strong, that's almost always a directory gap, regardless of what Sol's brand site reversal suggests.