GPT-5.6 Luna Is Now ChatGPT's Free Default — What That Changes for AI Visibility

On 6 August 2026, OpenAI announced two ChatGPT changes that look like product polish and are, for AEO, a measurement event.

GPT-5.6 Luna became the default model for Free and Go users that week. Unlimited text chats and a Think button for harder questions were scheduled to follow the next week, still on Luna. Plus and Pro users received a retuned GPT-5.6 Sol for everyday chat, plus a reasoning-effort slider. Work, Codex, and the API were explicitly left unchanged.

That split matters more than the press-release tone. ChatGPT reports 900 million+ weekly active users. Most of those people are not on Plus. If your prompt library only hits Sol — or an unnamed "ChatGPT" that your tool does not version — you are measuring a paid minority and reporting it as the market.

What actually shipped

OpenAI's internal eval, cited in coverage of the announcement, found responses with at least one factual error were 68% less common with the new Sol and 62% less common with Luna versus GPT-5.5 Instant on financial, medical, and legal prompts. Treat those as vendor-reported, not independent citation research — but they are the numbers OpenAI chose to lead with.

Limits still apply to file uploads, images, and other tools. "Unlimited" is text chat, not the whole product.

Why a cheaper default changes retrieval, not just chat quality

We covered the GPT-5.6 family at preview in What GPT-5.6 Changes About AI Search Visibility: cheaper frontier models tend to afford more retrieval and more reasoning per query, not less.

Luna as the mass-market default extends that logic to the people who never pay:

1. More ChatGPT Search sessions from free users. Unlimited text removes the "I already used my messages" brake. Category prompts that used to live on Google now have a lower-friction ChatGPT path.
2. A different reasoning budget than Sol. Free users get a Think button on Luna, not the Sol slider. Answers on the default path will be faster and thinner unless the user opts into more thought.
3. Paid and free answers can diverge. If Sol is retuned for "focused" everyday chat while Luna is the workhorse default, citation stacks on the same prompt will not automatically match across tiers.

None of that is a reason to panic-rewrite your site. It is a reason to label which ChatGPT surface you measured.

What to do this week

Re-baseline ChatGPT, not "GPT-5.6." Run the same funnel-weighted prompt set on:

• Free/Go default (Luna, with web search on)
• Plus/Pro Sol at default effort
• Plus/Pro Sol at a documented higher effort, if that is how your buyers work

Log model name, plan, retrieval on/off, date, and persistence band. A mention-rate swing that appears only on Luna is a free-tier story. A swing that appears on both is a retrieval-policy story.

Do not blend tiers into one ChatGPT score. IAB's directional vs decision-grade split still applies: a handful of Sol screenshots is exploratory, not a program. See Measuring AI Visibility.

Watch the week after the default change. Retrieval policy often moves with model rollouts even when OpenAI does not announce a search change. The site: fan-out jump and Reddit citation cliff that trackers documented days later (covered separately) are exactly the kind of second-order effect a default-model swap can precede.

What did not change

• Bing + OAI-SearchBot remains the ChatGPT Search backend. Google Search Console still does not proxy this path.
• Blocking GPTBot while allowing OAI-SearchBot is still the training-vs-retrieval split.
• Owned pages still lose most citations to earned media. A new default model does not make your homepage the primary evidence node.

How this collides with "one ChatGPT rank"

Agency reporting that still says "we appear in ChatGPT" without a model and plan is now actively misleading. A Free user on Luna with web search, a Plus user on retuned Sol at low effort, and a Plus user who dragged the slider to deep reasoning are three retrieval budgets. They will not cite the same three URLs.

If you sell retainers on screenshots, pick the tier that matches the client's buyers. Consumer apps skew Free/Go. Seats that live in ChatGPT Team/Enterprise may never see this Luna default at all — OpenAI scoped the change to Free and Go, and left Work unchanged.

FAQ

Is Luna "worse" for citations than Sol?

Unknown from public citation panels. OpenAI's pitch is fewer factual errors vs GPT-5.5 Instant, not a source-mix study. Measure both.

Should we block GPTBot now that Luna is free?

Training-bot policy is separate from SearchBot eligibility. Unlimited free chats do not change OAI-SearchBot. See training vs search crawlers.

Does unlimited chat mean unlimited Search?

OpenAI limited the "unlimited" claim to text chats. Search, uploads, and images can still be capped. Log whether live retrieval was on.

How Obsurfable fits

Obsurfable is built for this exact failure mode: the same prompt library, run on a schedule, with engine- and model-aware logs instead of a blended "ChatGPT rank." When OpenAI changes the default the majority of users hit, you re-baseline the prompt set — you do not invent a new KPI.

Bottom line

The ChatGPT most buyers use is now Luna, not the flagship. Treat 6–13 August 2026 as a forced re-baseline for ChatGPT Search. Measure the default people actually get, disclose the tier, and keep Sol as a paid-path column — not as a substitute for the market.