What exactly did OpenAI change on Aug. 26, 2026?

OpenAI removed o3 from ChatGPT entirely, according to its Model Release Notes page in the Help Center. The change closed a 90-day sunset window that had already been running, and it left GPT-5.6, offered as Sol, Terra, and Luna, as the full default model lineup across Free, Go, Plus, and Pro.

Does this affect the historical ChatGPT data an AI SEO tool already collected?

No. Data collected before Aug. 26, 2026 still reflects the model that was live at the time. The issue is comparison, not accuracy. A trend line that starts before the retirement and ends after it mixes two model versions in one chart, so a reader cannot tell how much of any change came from the model swap instead of the tracked content.

Does the o3 retirement change how Perplexity, Google AI Overviews, Gemini, or Copilot answer?

Not directly. OpenAI's Model Release Notes cover ChatGPT's own default model, not the other four engines this directory tracks. Each of those runs on its own model and its own release schedule, so a swap inside ChatGPT does not automatically move a reading taken from Perplexity, Google AI Overviews, Gemini, or Copilot.

How often should a brand re-run its tracked-prompt baseline?

On a fixed calendar is not enough on its own. Re-run the baseline every time the underlying default model changes, since that is the event that can move an answer independent of anything the brand published. A monitoring tool that logs the model version behind each pull makes that trigger easy to spot instead of easy to miss.

Do I need to switch AI SEO tools because of this retirement?

No single tool is at fault here, and switching will not fix the underlying issue. Every tracker that reports on ChatGPT sits downstream of whatever model OpenAI runs that day. The useful question is whether your current tool logs the model version and flags a swap date on its own charts, not which vendor's name is on the dashboard.

OpenAI Retired o3: Why Your AI SEO Tool's ChatGPT Baseline Just Reset

Last updated September 2026.

As of Aug. 26, 2026, o3 is gone from ChatGPT. OpenAI logged the change on its Model Release Notes page in the Help Center: the model’s 90-day sunset window closed that day, and GPT-5.6, running under the names Sol, Terra, and Luna, is now the complete default lineup across Free, Go, Plus, and Pro.

That is a routine platform update from OpenAI’s side. From the seat of anyone reading an AI SEO tool’s citation-rate chart, it is not routine at all. A default model is the version that answers a prompt when nobody selects a different one, and it is also the exact version every AI visibility tracker is measuring whenever its dashboard just says “ChatGPT.” Swap the model, and the sources it cites, the tone it uses, and the brands it names can all shift, with no change at all to the content sitting behind them.

What actually changed on Aug. 26, 2026

OpenAI began sunsetting o3 roughly 90 days before this date, giving developers and heavy users a runway to move off it. That runway closed on Aug. 26, 2026. o3 no longer answers inside ChatGPT, on any plan, free or paid. GPT-5.6 was already the default before the sunset finished, so the practical effect of this specific date is narrower than it sounds: it removes the last legacy option a user could still pick, and it closes out the changeover to the new model lineup that had been running in parallel for months.

For anyone tracking AI visibility, the date still matters. It is the point after which every ChatGPT answer in a citation report comes from the GPT-5.6 lineup and nothing else.

5 checks for anyone comparing “ChatGPT” charts across this date

Mark the swap date on the chart itself. A line labeled only “ChatGPT visibility, Q3 2026” hides the fact that the model behind it changed mid-quarter. Add a marker at Aug. 26, 2026, the same way you would mark a site migration or a major content push.
Re-run the baseline instead of stretching the old one. A tracked-prompt baseline set before the retirement was measuring o3-influenced answers, at least in part. Re-run it fresh after the changeover so later comparisons rest on GPT-5.6 data alone.
Separate a model swap from a content edit. If a brand’s citation rate moved the same week as this retirement, check the date before crediting or blaming a page update. The two events can land close together by coincidence, and only one of them involves anything your team published.
Cross-check against Perplexity, Google AI Overviews, Gemini, and Microsoft Copilot. Those four engines run on their own models and their own schedules, so they did not change on Aug. 26, 2026. If a visibility drop shows up on ChatGPT alone and not on the other four, that pattern points toward the model swap, not a site-wide problem.
Ask your tracking vendor how it dates its ChatGPT pulls. Some tools log the model version behind every answer; others just log the engine name. Knowing which one you have decides whether you can spot the next swap yourself or have to wait for the vendor to flag it.

Before and after: why straddling the date breaks a trend line

A chart that runs from July through September 2026 without a marker treats o3-era answers and GPT-5.6-era answers as one continuous series. They are not. Split any report that crosses Aug. 26, 2026 into a clean “before” period and a clean “after” period, and treat the “after” period as the true starting point for anything you plan to track going forward. A single averaged line across the swap will understate how sharp the actual change was in either direction.

What this means for the tools that track it

This directory covers several tools built to watch ChatGPT on an ongoing schedule, including Temso, Profound, Peec AI, AthenaHQ, and Otterly.AI. None of them controls when OpenAI changes its default model, and none of them is better or worse off than the others because of this particular retirement. What varies between them is whether the chart in front of you shows the model version behind each data point or just the engine name. That distinction is what turns a confusing dip into an explained one.

Before you trust the next citation-rate chart you open, check whether it survives this date. If the line runs straight through Aug. 26, 2026 with no marker, re-run the underlying baseline before you draw a conclusion from it. Compare tools that document their data-collection dates on the AI SEO tools index.