What model powers ChatGPT?

OpenAI’s assistant — the most-used AI chat app in the world.

As of July 2026, the free version of ChatGPT answers with GPT-5.5 Instant — OpenAI’s default model since 5 May 2026, when it replaced GPT-5.3 Instant. Paid tiers can reach the newer GPT-5.6 family in reasoning mode, but GPT-5.5 Instant stays the default for everyday chat. Discoverably reads the model from ChatGPT’s own answers, so this is what your customers actually get.

Current model
GPT-5.5 Instant
Evidence
Observed
As of
May 2026
What we capture
chatgpt.com, the free logged-out experience

Discoverably reads this model from the engine’s own answers, so it reflects what the free experience actually returns — not only what the vendor last announced.

How ChatGPT picks its model

ChatGPT auto-selects a model for you; the free experience is not a menu. Since 5 May 2026 that default has been GPT-5.5 Instant, which OpenAI rolled out to every free user in place of GPT-5.3 Instant. When GPT-5.6 reached general availability on 9 July 2026, it landed on paid tiers as a reasoning-mode option (GPT-5.6 Sol) — GPT-5.5 Instant stayed the default for ordinary questions, so the logged-out answer your customers see is still GPT-5.5 Instant.

OpenAI can change this default at any time, without notice. Discoverably follows it automatically rather than pinning a version, because the point is to measure what real people are shown — not a model we chose.

Model history

Every family-level change to the model behind ChatGPT, newest first, each linked to its primary source.

GPT-5.5 Instant
Model changeObserved

GPT-5.5 Instant becomes the default model for all free ChatGPT users, replacing GPT-5.3 Instant.

OpenAI ↗ (opens in a new tab)
GPT-5.6 (paid, reasoning mode)
Model changeVendor-stated

GPT-5.6 reaches general availability on paid ChatGPT tiers as a reasoning-mode option; GPT-5.5 Instant stays the default for everyday chat.

OpenAI model release notes ↗ (opens in a new tab)
  1. GPT-5.6 (paid, reasoning mode)

    GPT-5.6 reaches general availability on paid ChatGPT tiers as a reasoning-mode option; GPT-5.5 Instant stays the default for everyday chat.

    OpenAI model release notes ↗ (opens in a new tab)
  2. GPT-5.5 Instant

    GPT-5.5 Instant becomes the default model for all free ChatGPT users, replacing GPT-5.3 Instant.

    OpenAI ↗ (opens in a new tab)

What decides whether ChatGPT names your brand

ChatGPT answers in two regimes — from training (parametric), or from a live web search it runs itself — and which one a question lands in changes what drives brand visibility.

How answers form & what drives brand mentions

Regime 1 — parametric (no search). OUR INFERENCE (mechanism VENDOR-DOCUMENTED) For questions the model judges answerable from training, ChatGPT answers from learned parameters. OpenAI documents that its models learn from public internet content, licensed data, and user/trainer input, storing patterns in weights rather than a browsable copy. Consequence: a brand’s pre-existing prominence and consistency in the training corpus is the dominant discovery mechanism, and nothing published today changes an already-trained model.

Regime 2 — search-grounded. VENDOR-DOCUMENTED ChatGPT decides on its own when current web information would help, or the user forces search; it then “rewrites your query into one or more targeted queries” sent to search providers, and OpenAI operates its own crawler, OAI-SearchBot, whose index surfaces and links sites in ChatGPT search. The blend of third-party providers vs OpenAI’s own index is undisclosed. OUR INFERENCE: retrievability across the rewritten queries, not just the literal prompt, is what matters when search fires.

The crawler gate. VENDOR-DOCUMENTED Three separately controllable agents: GPTBot (training), OAI-SearchBot (search index; not used for training), ChatGPT-User (live user-initiated fetches). A site that blocks OAI-SearchBot will not appear in ChatGPT search answers, though it may still show as a navigational link; robots changes propagate in ~24h.

Commerce is a distinct path. VENDOR-DOCUMENTED For shopping, ranking can consider relevance, first/third-party structured metadata, price, reviews, availability, and maker/primary-seller status; Shopify catalog data is integrated and merchants can provide direct feeds. Product results are not ads and not ordered by partnership status.

What retrieval favors. INDEPENDENT STUDY Profound’s analysis of 680M+ citations found Wikipedia the single most-cited ChatGPT domain (~7.8%), with Reddit, Forbes, and G2 following — treat the exact share as a snapshot of its index window. Ahrefs’ 75,000-brand study (2025-12) found ChatGPT visibility correlates only weakly with classic authority metrics (Domain Rating ~0.27), the weakest of the Google-plus-ChatGPT set. OUR INFERENCE: ChatGPT is the most accessible of the six engines for smaller brands, because domain-authority hierarchies transfer least — content that directly answers the rewritten sub-queries can win citations without a top-ranked page.

Factor weights

Weight and confidence are our editorial estimate of each factor’s influence, and of the evidence for its direction — not a vendor’s ranking weight. No engine publishes a brand-selection formula.

FactorWeightConfidenceBasis
OAI-SearchBot crawl eligibility (binary gate for owned-page citations)High (binary gate)HighVENDOR-DOCUMENTED
Retrievability/relevance across ChatGPT’s rewritten queries when search firesHigh (conditional)HighVENDOR-DOCUMENTED (that a blend exists); weighting OUR INFERENCE
Learned brand–topic association when search does not fireHigh (conditional)MedOUR INFERENCE (mechanism vendor-documented; brand weighting undisclosed)
Presence on heavily-cited reference/community surfaces (Wikipedia, Reddit, review sites)HighMedINDEPENDENT STUDY (Profound 2025); volatile over time
Content extractability — direct answers to conversational sub-queries, clear structureMed-HighMedOUR INFERENCE; consistent with the KDD-2024 GEO benchmark
Product metadata, direct feeds, price/availability/reviews, maker status (commerce only)High (commerce only)HighVENDOR-DOCUMENTED
Freshness for time-sensitive searched queriesMedMedOUR INFERENCE from documented search-trigger behavior
Classic domain authority (DR, backlinks)Low-MedMedINDEPENDENT STUDY (Ahrefs 2025-12: weak ChatGPT correlations)
Generic schema markup (outside shopping)LowLowOUR INFERENCE; no vendor doc that ChatGPT search consumes schema

What you can do (ranked)

  1. Make owned pages eligible for ChatGPT searchengine-specific · highest-confidenceVENDOR-DOCUMENTED

    Allow OAI-SearchBot in robots.txt and at the CDN/WAF (bot-mitigation false positives are a common cause of missing citations), return indexable content, re-test after ~24h. GPTBot (training) is a separate choice, not required for search visibility.

  2. Build pages that resolve the likely rewritten queriescross-engineOUR INFERENCE

    A category page alone under-covers a prompt ChatGPT fans into product, use-case, comparison, proof, constraint, and local sub-queries. Publish self-contained evidence for those intents.

  3. Earn category-specific corroboration where retrieval already lookscross-engineINDEPENDENT STUDY + OUR INFERENCE

    Prioritize credible reviews, trade press, community, and video actually retrieved for the target prompts, not indiscriminate mention volume. SE Ranking (129k domains) found domains present on review platforms (G2, Capterra, Trustpilot) averaged ~2.5–3.5× the ChatGPT citations of absent domains — an association, not proof of cause.

  4. Keep claim-bearing pages current and machine-readablecross-engineOUR INFERENCE

    Dates, units, geography, product names, and qualifiers in HTML text near the evidence, with stable canonical URLs. No OpenAI source promises a schema or “LLM-friendly” ranking bonus.

  5. Use the merchant path for product discoverycommerce-onlyVENDOR-DOCUMENTED

    Accurate availability/price/reviews/metadata; a direct feed if eligible; verify Shopify catalog data.

  6. Measure search and non-search prompts separatelyengine-specificOUR INFERENCE

    An intervention that improves citations cannot be assumed to improve parametric answers.

Grounding, citation & locality

VENDOR-DOCUMENTED Searched answers can carry inline citations plus a Sources panel (cited + related links); the API can return a fuller source list than the answer visibly cites. Parametric answers carry no retrieval citations — any source-like text is generated, so parametric mentions cannot be earned with citation tactics. Locality: ChatGPT may use IP-derived general area (country/region/city); Memory can influence the rewritten query; precise device location is opt-in.

INDEPENDENT STUDY (do not treat as a current rate) A peer-reviewed audit of ~14,000 search-enabled LMArena interactions found large visited-vs-cited gaps across systems.

Sources

Three steps to get started.

A couple of hours a month is enough to make real progress.

Enter your website.

We look up your business and the questions your customers actually ask AI.

Review what we found.

Confirm your business, your questions and your competitors. Then we ask the live apps your customers actually use: ChatGPT, Gemini, Google AI Overviews, AI Mode, Copilot and Perplexity. Not the developer API, and not the model's training data. Both give a different answer from the one your customer sees.

Get your free report.

Where you appear, who wins instead, and how AI describes you, in plain English with clear next steps. A one-time snapshot to start from.

See what AI says about you.

Get a free report: no sign-up, no credit card, results in minutes.  ·  See a sample report ›