ChatGPT: what each inline product card isChatGPT shows brands, companies, software, places and ideas in the same cards as shop products. Each product in
inlineProducts now has kind and entityCategory. kind is product for a shop product, entity for a named entity such as a brand, and row for any other card. entityCategory is the category that ChatGPT gives to the entity in the card title, such as brand or tv_show, or null. To keep only shop products, filter on kind: "product".price now holds only a price. Before, a card for a business could have its rating or address in price. That line now stays in subtitle or description.ChatGPT: call-to-action links on product cardsA gpt-6 product card can show a call-to-action line, such as “Acheter chez TennisPro”. Each product in
inlineProducts now has ctaLinks, with the label and URL of each merchant link on that line. Use it to tell these links apart from citation chips (citationPillIds) and product sheet offers (offers). A product with no call-to-action line has ctaLinks: []. url does not change.ChatGPT: answers kept when ChatGPT does not search or citeThe ChatGPT endpoint forces web search, and it used to fail any attempt that showed no sign of search. Two things changed:
- An answer where ChatGPT searched but cited nothing now succeeds, with empty
sourcesandcitationPills. - If ChatGPT does not search on attempts 1 and 2, cloro keeps the answer from the third attempt on, searched or not.
sources, citationPills, and searchQueries even without disableWebSearch. Retrying these prompts rarely changes the result. See Why are sources and citationPills empty when web search is forced?ChatGPT: product rationale and reviewsSet
include.productDetails: true with include.shopping: true on the ChatGPT endpoint to get the “What to know” rationale and the “What people are saying” reviews that ChatGPT shows when you open a product. They come back as productDetails on each product in shoppingCards and inlineProducts, and on each entity listing in entities[n].product. This adds +2 credits to the request, on top of the +2 credits for include.shopping. See product details.ChatGPT: all offers on shopping productsoffers on shopping-card and inline products now has every offer ChatGPT shows under “Compare offers”. Before, it had only the first one. Answers from ChatGPT’s mobile interface now have offers and url too, as desktop answers already did.ChatGPT: gpt-6 product cards are in inlineProductsEach gpt-6 product card is now an inline product. A card with a listing has it in inlineProducts, not in entities[n].product. An entity keeps its listing only when it is outside a card. A card with no listing is an inline product too: its product comes from the card (title, link, price text, badge, subtitle, description, “Label: value” lines such as “Best for: …” in attributes, and image), with no id and no offers. Each card also has citationPillIds: the citationPillId of each citation chip in the card, so you can find in result.citationPills which sources back each product.ChatGPT: entities now have their product listingWhen ChatGPT attaches a product listing to an entity, the ChatGPT endpoint now returns it as
entities[n].product: the price, the URL and the merchant offers, in the same structure as a shopping card product. It needs include.shopping: true, at no extra cost over that add-on.ChatGPT: multi-turn conversationsThe ChatGPT endpoint now accepts Each prompt after the first costs +1 credit. A
prompt as an array of up to 5 prompts. ChatGPT answers them in order, in the same conversation, and result is an array with one response for each prompt. Use it to measure what ChatGPT says, cites and advertises after the first answer.prompt string works as before and returns one object.ChatGPT currently shows ads only in answers to a follow-up prompt, not in the answer to the first prompt of a conversation. To get result.ads, send two or more prompts with include.ads: true and legacy: true, and read the ads from the second response or a later one. Follow-ups on the default mobile-web interface return ads far less often.ChatGPT: the follow-up form is now in the responseChatGPT now ends many answers with a form that refines the prompt, such as “What’s your budget?” with a list of ranges. The ChatGPT endpoint returns it as
result.followUpForm, with no flag and at no extra cost: the questions, their options, the one ChatGPT selects by default, and the follow-up prompt it would send.followUpForm is null when the answer has no form. The form text no longer shows in text and markdown.ChatGPT: Czech Republic (
CZ) is supported againThe ChatGPT endpoint accepts country: "CZ" again. We removed it on 30th June 2026 because ChatGPT showed a login wall to anonymous users from Czech IPs. ChatGPT no longer shows that wall, and answers for CZ are localized to the Czech Republic. The /v1/countries endpoint lists CZ under model=chatgpt again.ChatGPT: opt out of forced web search with
disableWebSearchThe ChatGPT endpoint forces ChatGPT’s web search on every request. Set disableWebSearch: true and cloro returns the answer ChatGPT gives on its own, searching only when it decides to. Most prompts then come back without search, so result.sources, result.citationPills, and result.searchQueries are often empty, and the answer text can differ from the search-forced one. Use it to see what ChatGPT says without search, and keep the default to measure citations. The cost is unchanged, and requests that do not send the flag behave exactly as before.ChatGPT: request the legacy interface with
legacyThe ChatGPT endpoint accepts a legacy boolean. Set it to true and cloro asks ChatGPT for its legacy desktop interface instead of the mobile-web one that now serves almost all traffic. That interface streams the answer as an event stream, so result.model, result.searchQueries, and result.mapSearchQueries come back filled in rather than empty. Unset, requests behave exactly as before.ChatGPT: the mobile-web format now serves 98% of responsesOpenAI completed the rollout of the mobile-web response format on 25 August 2026. It served approximately 10% of successful ChatGPT responses through early August, approximately 50% between 18 and 24 August, and 98% from 25 August.Nothing changed in the cloro API. The effects are the ones in the 21st July 2026 entry below: OpenAI selects the format per request on their side. cloro replicates a real user and does not force a particular interface, so we cannot select the event-stream format for you.
result.model, result.searchQueries, and result.mapSearchQueries come back empty, result.sources holds fewer entries, and rawResponse uses the HTML-fragment shape. The answer text / markdown and inline citationPills are unaffected.An empty field does not mean the field is gone. Providers have restored this kind of metadata later, so keep your parsing for
result.model, result.searchQueries, and result.mapSearchQueries in place.ChatGPT: new mobile-web response formatOpenAI now serves a share of logged-out ChatGPT traffic through a new mobile-web format that streams the answer as HTML fragments rather than the usual event stream. On responses served this way:
rawResponse(when you opt in) has a different, HTML-fragment shape alongside the existing event-stream shape.result.model,result.searchQueries, andresult.mapSearchQueriescome back empty — the mobile format carries none of the stream metadata these are parsed from.result.sourcesis shorter; the format exposes fewer full source entries.- The answer
text/markdownand inlinecitationPillsare unaffected.
ChatGPT:
visibility, adsResponseIndex, and rendered on adsEach ad in result.ads (set include.ads: true) now carries three fields that separate what OpenAI served from what the user actually saw. OpenAI serves candidate ads but the interface renders at most one; previously result.ads listed the served candidates with no way to tell which was shown. rendered is true only for the ad displayed as a visible card, so you can count real impressions instead of the array length. visibility exposes OpenAI’s serving status (allowed, or hidden when an ad was filtered), and adsResponseIndex gives the ad’s served slot position. The fields ride the existing include.ads flag, so there is no new parameter and no extra credit cost. See the ChatGPT ads schema.ChatGPT: new
mapSearchQueries fieldThe ChatGPT endpoint now exposes a mapSearchQueries array on result when you set include.searchQueries: true. When ChatGPT renders a map block for a local-intent prompt (e.g. “best coffee shops in San Francisco”, “plumbers near me”), it fires a separate query at its maps tool to gather the businesses behind result.map. That query previously lived only in the raw stream. It now has a structured home, distinct from result.searchQueries (the web-search fan-out), which the two can diverge on for the same prompt. The field rides the existing include.searchQueries flag, so there is no new parameter and no extra credit cost. It is additive and omitted when no map appears in the response. See the ChatGPT response schema.Czech Republic (CZ) targeting disabled in ChatGPTWe’ve removed
country: "CZ" support for the ChatGPT endpoint. ChatGPT no longer allows anonymous access from Czech IPs — every request is met with a login wall, which scrapers cannot pass through. Requests with country: "CZ" to /v1/chatgpt will now return an unsupported-country error instead of silently failing.State-level geo-targeting for AI endpointsYou can now target a specific US state when querying ChatGPT, Copilot, Perplexity, Gemini, or Grok. Pass the USPS two-letter code as the State targeting costs +2 credits on top of the base cost and any other add-ons. Use
state parameter alongside country: "US" and cloro routes the request through a proxy in that state — giving you localized results at the sub-country level.GET /v1/states?country=US to retrieve the full list of supported state codes. Google Search and AI Mode are not affected — those endpoints continue to use location / uule for sub-country targeting.Inline products now return offers correctly for ChatGPTWe fixed a bug where ChatGPT inline products were missing pricing and merchant data. Inline products now include the full offer payload (current price, multi-merchant offers from retailers like Amazon and John Lewis, product images, and aggregated ratings) fetched from ChatGPT’s product update API.Alongside this fix, shopping cards and inline products move to an opt-in model on the ChatGPT endpoint. Set Learn more in our ChatGPT shopping cards documentation and inline products documentation.
include.shopping: true to include them in your response:- Pricing and offers: Current price and multi-merchant offers with availability
- Product images: Image URLs for product display
- Ratings: Aggregated rating scores and review counts
Opt-in with additional credit costShopping cards and inline products are disabled by default. Set
include.shopping: true to enable them. Enabling shopping, rawResponse, searchQueries, or ads (or any combination) adds +2 credits to the base cost.The credit surcharge reflects increased extraction complexity. OpenAI now serves product pricing, offers, and merchant data through a separate endpoint, so cloro performs additional fetches per product to enrich the response.Breaking change: Shopping cards and inline products were previously included automatically in ChatGPT responses. They now require
include.shopping: true in your request. The details object on inline products (rationale and themed reviews) is no longer returned.More business data in ChatGPT map entriesChatGPT map entries now return mapped business data from providers (Yelp, Google Business) with 47 structured fields, replacing the previous 10-field structure.What changed:Learn more in our ChatGPT map documentation.
- Mapped data: Map entries now include all available business fields from ChatGPT’s providers (Yelp, Google Business, Yelp-feed)
- 47 mapped fields: Access to detailed location data, operating hours, attributes, reviews, images, and provider-specific metadata
- Structured schema: Properly typed fields with consistent camelCase naming conventions
- Field name updates:
phone_number→phone,country→countryCode,review_count→reviewCount - Null-safe: All fields except
nameandpositionare optional
- More location data: Full address breakdown with coordinates (
latitude/longitude), distance from user (distanceMeters), and human-readable location strings - Hours: Weekly schedule with next opening times (
nextOpenHour) and special hours (specialHours) - Attributes: Provider-specific details like parking options, WiFi availability, accepted payment methods, and accessibility features
- Images: Multiple image URLs (
imageUrls) with provider-hosted images (providerImages) - Metadata: Business claimed status (
isClaimed), opening/closing dates (dateOpened/dateClosed), popularity scores (popularityScore), and cache indicators (fromCache)
Breaking change: Field names have changed to camelCase. Examples:
review_count → reviewCount, country_code → countryCode, is_open → isOpen. All fields except name and position are now optional.Ad extraction for ChatGPTChatGPT responses now support structured ad extraction when advertising content is displayed. This opt-in feature lets you monitor ads shown in ChatGPT responses.Features:
Opt-in feature with additional costAd extraction is disabled by default. Set
include.ads: true to extract ads from ChatGPT responses. Enabling ads, rawResponse, or searchQueries (or any combination) adds +2 credits to the base cost.- Brand information: Advertiser brand name, URL, and favicon
- Carousel cards: Multiple promotional cards with titles, descriptions, images, and destination URLs
- Attribution tracking: All URLs include ChatGPT attribution parameters
- Structured data: JSON structure
Inline products for ChatGPTChatGPT responses now include inline products: individual product references that appear embedded in the response text, separate from shopping cards. This feature supports OpenAI’s new product discovery capabilities announced in ChatGPT.What’s new:
Model availability: Inline products currently appear when ChatGPT uses the
gpt-5-3 model. ChatGPT automatically selects which model to use for each request.- Inline product extraction: Individual products referenced in comparison tables, featured recommendations, or inline mentions
- Rendering hints: Display guidance with
render_asfield (“inline”, “hero”, or “block”) - Cite references: Each product includes a unique cite ID for cross-referencing with text
Improved ChatGPT source metadataChatGPT sources now include enhanced metadata for better source attribution:
datePublished: Publication date of the source article (e.g., “May 22, 2025”)label: Now contains the article title instead of the domain namedescription: Now contains a content snippet instead of the combined date and title
ChatGPT geographical expansionChatGPT is now available in 200 countries worldwide.The countries endpoint remains the authoritative source for current country availability. Always query with
model=chatgpt to verify supported locations before deploying.Isolation of ChatGPT citation dataWe’ve improved how source information is structured and exposed in ChatGPT responses:Citation pills: Inline citations that appear within ChatGPT responses where specific sources are referenced. Each citation pill includes:
- URL: Direct link to the cited source
- Label: Title or label of the citation
- Description: Summary of the cited content
- Domain: Source domain (e.g., “example.com”)
- Date published: ISO 8601 date string when the source was published
- Citation pill ID: Unique identifier for the citation
sources array now includes a footnote boolean field that indicates whether a source appears in the main window footnote (i.e., the sources pill), to distinguish between primary and secondary sources.Previously, the citation and source pill data was only accessible by parsing the markdown response. Now you can access it directly through structured JSON objects without needing to parse markdown links.No additional parameters required. Citation pills and the footnote field are automatically included when available in ChatGPT responses.Learn more in our ChatGPT documentation.Map entries for ChatGPTChatGPT responses now automatically include structured map entries when business or place information is detected. Use cases:
- Local business monitoring: Track restaurants, stores, and service providers
- Location intelligence: Extract ratings, reviews, and contact details
- Competitive analysis: Monitor local businesses across different regions
- Directory building: Create structured business listings
- Business information: Name, category, description
- Social proof: Rating score, review count
- Contact details: Website URL, phone number
- Location data: Position/ranking in results
- Navigation: Directions URL (when available)
Entity extraction for ChatGPTChatGPT responses now include automatic entity extraction when specific items, products, brands, or concepts are identified. Use cases:
- Product recognition: Identify and extract product mentions
- Brand tracking: Monitor brand references across ChatGPT responses
- Concept analysis: Extract named entities for semantic analysis
- Content classification: Categorize response content based on extracted entities
- Type: Entity type identifier (e.g., “product”, “software”)
- Name: Entity name or title (e.g., “adidas Grand Court Lo”, “Nike Dunk Low Retro SE”)
Shopping cards for ChatGPTChatGPT responses now automatically include structured shopping cards when product information is detected. Use cases:
- Product monitoring: Track prices, ratings, and availability over time
- Price tracking: Monitor price changes across different regions
- Competitive intelligence: Extract structured product data for analysis
- Commerce integration: Build price comparison and recommendation systems
- Product information: Name, brand, description, specifications
- Pricing data: Current price, original price, discount information
- Commercial details: Multiple merchant offers, availability, checkout options
- Media assets: Product images, checkout-specific images
- Ratings: Ratings, review counts, rating citations with sources
- Offer details: Merchant-specific pricing, promotional tags, shipping costs
- Multi-offer products: Compare different merchants for the same product
- Price breakdown: Pricing including base, shipping, tax, and total
- Promotional content: Tags, tooltips, and special offers
- Media: Multiple product images and checkout assets
- Availability tracking: Real-time stock and checkout status
New dedicated endpointsEach AI model now has its own dedicated endpoint for better performance and clearer documentation:
- Extract Google AI Overview (via
POST /v1/monitor/googlewithinclude.aioverview) - Extract Microsoft Copilot (
POST /v1/monitor/copilot) - Extract Perplexity (
POST /v1/monitor/perplexity) - Extract Google AI Mode (
POST /v1/monitor/aimode) - Extract ChatGPT (
POST /v1/monitor/chatgpt)
Added Learn more in our API Reference.
include.rawResponse parameterYou can now access the underlying AI provider payload in monitoring responses (ChatGPT-only). Enable it with "include.rawResponse": true to receive the full provider response alongside cloro’s structured output.