OpenAI rebuilt ChatGPT's search tool from the ground up. Not an incremental update. Not a patch. A complete rewrite of how the platform formulates, sends, and processes search queries. And what surfaced tells us more about the future of SEO than any Google update from the past 12 months.
A recent analysis on Search Engine Journal examined the new query format and uncovered a revealing pattern: 84 Reddit threads entered the retrieval pool. Zero were cited in the final answer. This raises a question we test daily in our SEO work: if AI reads your content but doesn't cite you, what's visibility worth?
Four days, a completely new language
OpenAI replaced its JSON query format with a pipe-delimited language: fast|query|30|domain.com. The first value indicates the search type. The second is the query itself. The third sets a freshness window in days. The fourth targets a specific domain (when applicable). The entire switch happened in four days. Not months. Four days.
What's truly interesting isn't the format itself but what it reveals about how ChatGPT "thinks" before it searches. Each content type gets a different freshness window. Stock prices get a 2-day window. Commercial products get 30 days. Earnings guidance gets 90 days. And Reddit searches? Up to 3,650 days, roughly 10 years.
The logic is straightforward: ChatGPT first classifies what kind of answer it needs, then searches. It's not a generic query. It's a search with context, intent, and category-specific rules. This fundamentally changes which content gets a chance to be found.
For an e-commerce site, this means product pages compete within a 30-day retrieval window. If you haven't updated descriptions or pricing in the past month, ChatGPT treats the information as stale and looks elsewhere. For a marketing blog like ours, the window is likely similar, which explains why older-but-accurate articles see declining traffic from AI search. Freshness isn't just about dates anymore. It's baked into the query architecture itself.
Reddit: the input nobody sees
This is where things get interesting from an SEO perspective. Reddit's share of ChatGPT citations dropped from roughly 3.83% to under 1% in August 2026. A collapse of nearly 75%. Yet Reddit continues to be pulled from the retrieval pool. ChatGPT still queries Reddit, still processes responses, but no longer attributes them publicly.
Reddit has become what analysts call an "invisible input": a source that feeds the verdict without getting credit. Reddit opinions shape how ChatGPT frames an answer, but the end user never sees a link to the original thread.
The implication is direct. If Reddit, one of the most cited sources on the web, can become invisible in ChatGPT's output while remaining part of its input, what guarantee do you have that your own content won't follow the same path? We've previously analyzed how much real communities matter in AI search. Now the data shows that value manifests differently than we expected: not through direct citation, but through indirect influence.
What we see in our client work confirms this pattern. ChatGPT referral traffic doesn't grow proportionally with indexed content volume. Sites with articles frequently cited by other publications receive AI search visits, but sites with useful information but no editorial authority appear less and less in explicit citations. Their content is processed, but it doesn't get credited.
Widgets and the complete disappearance of citations
The new ChatGPT Search doesn't run a single search type. The data reveals at least five specialized verticals: fast (general web), product (product catalogs), business (places and services), image (visual search), and genui_run (interactive widgets).
That last category should concern any SEO practitioner. When ChatGPT responds with a financial chart, a sports schedule, or a built-in calculator, citations cease to exist. Zero links, zero attribution. The answer is generated entirely within the interface, with no reference to the data source.
It's a natural extension of what we observed with the shift from ranking to citation: if the new "position 1" was being cited in the AI answer, widget responses eliminate even that possibility. You're no longer competing for a position. You're competing for the position to exist at all.
As ChatGPT expands the query categories that receive widget answers (from finance to sports, from weather to currency conversions), the volume of queries where traditional visibility simply doesn't exist will keep growing.
What you can actually do about it
First recommendation: understand which vertical your content falls into. If you publish information that can be turned into a widget (financial data, sports scores, conversion tables), your probability of earning a ChatGPT citation drops to near zero. That doesn't mean your information isn't used. It means attribution disappears.
Second: update cadence matters more than you thought. If the freshness window for your category is 30 days but you update content quarterly, you're effectively invisible to ChatGPT 60% of the time. Align your publishing frequency with the relevant freshness window.
Third: focus on content that generates perspective, not just information. The formats that hold up best are those offering original analysis and context. They're hard to turn into widgets and hard to process without attribution. This is exactly what we argued when discussing why AI prefers citations over backlinks: content with a clear point of view has the highest chance of being explicitly cited.
Fourth: monitor differently. Google Search Console won't tell you if ChatGPT reads you without citing you. You need server logs that identify AI crawlers and tools that measure mentions, not just links. Classic referral traffic is no longer sufficient as a metric.
And one final recommendation, perhaps the most important: diversify your formats. Long-form articles with original analysis remain the best positioned for citation. But podcasts, interviews, and case studies with proprietary data also carry growing value. ChatGPT appears to prefer sources that add something that doesn't exist elsewhere. If everything you publish can be found on 50 other sites, you're the perfect candidate for invisible input status.
ChatGPT just showed its hand. Not the full hand, but enough to understand the rules of the game. And the rules are straightforward: not everything it reads will be cited, not everything it searches will be shown, and not everything it processes will be attributed. The question is no longer whether AI can find you. It's whether, after finding you, it has any reason to mention you.
