What changed in ChatGPT search behavior in August 2026
What OpenAI documented for ChatGPT search in August 2026, what did not change, and which tools help you spot shifts in citations and visibility yourself.
If your ChatGPT visibility moved in August 2026 and you are looking for a single announcement that explains it, the documented record is quieter than you might expect. OpenAI's release notes and help pages show real product changes that month, but most are about what ChatGPT can do around search, not about how it picks and cites sources. This article separates what was documented from what was observed, and shows how to detect the rest on your own data.
The sourcing here is OpenAI's own Help Center release notes and the rewritten "Searching the web with ChatGPT" help page, read on 31 October 2026, plus a public change tracker that watches that page.
What OpenAI documented in August 2026
Restaurant reservations (release note dated 10 August). ChatGPT can now show live reservation times from OpenTable, Resy, and Yelp inside a response and let people book from the chat. The note says it is rolling out across all ChatGPT plans on mobile, web, and desktop, with OpenTable available globally, Resy in the US, and Yelp in the US and Canada. ChatGPT Work does not include it. For local and hospitality brands this is a new type of answer on a search-like request, one where listings from those three partners appear directly.
Model changes (same note). Plus and Pro users got an updated GPT-5.6 Sol with more reliable facts and a slider for how much thought ChatGPT applies. GPT-5.6 Luna was set to become the default for Free and Go users, with a Think button for harder questions. A model change can alter how answers read, so if your tracked results shifted in this window, it is a candidate cause. The note does not describe any change to how sources are chosen.
The web search help page was rewritten (around 20 August). The page was retitled "Searching the web with ChatGPT" and reorganized, with sections on availability, troubleshooting, reviewing sources, location, and restaurants. The availability list now names Free, Go, Plus, Pro, Business, Enterprise, and Edu, and says people who are not signed in can also use web search. A change tracker that diffed the page reported that the documented retrieval mechanics were unchanged: ChatGPT may search automatically when current information would help, and rewrites a question into one or more targeted queries. The page also now says to open cited sources and check that they support the answer.
Temporary chat controls (27 August). Users can choose personalized temporary chats that draw on memory, plugins, and custom instructions. This matters for anyone checking answers by hand, because your own chat history can shape what you see.
What was not documented
We found no release note that describes a change to ranking, source selection, or citation format in ChatGPT search in August 2026. If you saw a shift in who gets cited, the documentation does not explain it. Outside reporting did: Search Engine Journal published a piece on an August 2026 drop in Reddit citations, drawing on Promptwatch data. We do not have enough sourced detail to describe its size or cause here, so treat it as a sign to check your own Reddit-related citations.
How to find out what changed for you
Release notes tell you what OpenAI announced. Your own data tells you what changed for your brand. The method:
- Compare two periods on the same prompt set, one before and one after the date you noticed.
- Check whether mentions, citations, or both moved.
- Look at the query fan-outs ChatGPT generated, since rewritten queries decide which sources are retrieved.
- Check which domains gained or lost citations, particularly community sources.
- Rule out your own side: crawler errors, site changes, or a competitor publishing.
1. Promptwatch: built for this comparison
Promptwatch runs prompt panels daily per model, country, and language, so you have a baseline to compare against. Its A/B period comparison puts two windows side by side. Query fan-outs show what each model searched. Citation Trends classifies each citation by content type and source type and shows citation rate over time, along with first cited, ramp-up, peak, and decay half-life per page. Reddit, YouTube, and LinkedIn citation reports isolate community sources, and Agent Analytics (IP-verified crawler logs for GPTBot, OAI-SearchBot, and ChatGPT-User, among others) tells you whether a drop came from your side. Plans start free (Explore, ChatGPT only) and paid from $95/mo. Crawler logs begin on Professional ($245/mo).
2. Other tools and how fast they notice
AthenaHQ offers multi-engine tracking and an action queue, with credit-based usage. Scrunch AI refreshes weekly, so a mid-week shift can take days to show. Ahrefs Brand Radar retests on roughly monthly snapshot cycles, which is slow for a change that started on a specific day. Semrush AI Toolkit uses AI-generated approximations of user prompts.
Comparison table
| Tool | Refresh cadence documented | Period comparison | Crawler data |
|---|---|---|---|
| Promptwatch | Daily prompt panels | A/B period comparison | IP-verified logs |
| AthenaHQ | Not stated | Not stated | Not published |
| Scrunch AI | Weekly | Not stated | Not published |
| Ahrefs Brand Radar | Roughly monthly snapshots | Not stated | No |
The takeaway
August 2026 brought documented changes to what ChatGPT can do around search, notably restaurant bookings and model updates, plus a rewritten help page. It did not bring a documented change to source selection. If your numbers moved, test your own data against those dates rather than assume a cause, and keep a daily baseline so you can tell the next change from noise.