What We Can Learn From Evolving ChatGPT Fan-Out Queries

Aug 18, 2026 08:01 PM - 1 hour ago 5

Over the past fewer months, I’ve been paying adjacent attraction to really ChatGPT’s fan-out queries person been evolving, arsenic immoderate absorbing developments person taken spot that I judge are changing the value of ChatGPT’s responses. I’ve been moving connected this article for a while, but this 1 has been peculiarly difficult to write, because arsenic pinch each things successful AI search, the accusation changes much quickly than I tin decorativeness penning astir it. That’s decidedly been existent for really OpenAI appears to beryllium tweaking and refining its process of retrieving accusation via web hunt (RAG), and particularly for really heavy ChatGPT has started relying connected site: searches successful fan-out queries, perchance utilizing them to curate results from higher-quality sources.

The TL;DR: I deliberation OpenAI is utilizing fan-out queries, and the site: usability successful particular, arsenic 1 method of reducing spammy outputs successful their answers derived from net content. I judge it’s their effort to amended the value of retrieved sources while taking early steps to combat spam and low-quality accusation successful their results. It reminds maine of what Google has tested pinch E-E-A-T, but ChatGPT style.

I deliberation analyzing fan-out queries matters because ChatGPT is utilizing hunt engines to retrieve the results it uses to formulate an answer, which intends this is fundamentally an SEO problem. Every connection the exemplary chooses to put into a fan-out query, and each search operator it uses, tells america thing astir what the exemplary is looking for and wherever successful the hunt results it expects to find it. When it scopes a hunt pinch site:, adds the connection “official,” aliases points astatine a circumstantial subreddit, the exemplary is telling america what benignant of contented it believes will champion reply the user’s question. Those queries are the closest point we person to knowing why ChatGPT pulls successful the accusation that it does, and I deliberation location is simply a batch we tin study from unpacking them.

There are a fewer folks successful the manufacture who person done awesome activity sharing their findings connected the soul workings of ChatGPT: reference the earthy traffic, scraping the speech files, and pulling fan-outs retired of the API. Their datasets person started to converge connected the aforesaid findings, and this article combines the learnings from their investigation pinch my ain observations from watching fan-out behaviour complete time, utilizing a operation of Peec AI, Profound, the Resoneo plugin, FanoutFox, and Google Search Console. So I purpose to do 2 things pinch this article: First, laic retired what everyone has really found, successful 1 place, pinch the numbers attributed to whoever ran them. Then springiness you my publication connected what’s evolving and why I deliberation it’s important.

The Mechanics Of How ChatGPT Fan-Out Queries Work

Not each searches connected ChatGPT usage web search. Anything contained successful its training information tin beryllium answered quickly without utilizing RAG, and OpenAI’s free aliases cheaper models are more apt to trust connected training data to reply questions quickly, arsenic it costs them little money to generate. When the mobility requires up-to-date knowledge, ChatGPT will usage web hunt (retrieval-augmented procreation aliases RAG) to propulsion accusation from a assortment of sources, including outer hunt motor information and its ain soul scale (Labrador).

When you do inquire ChatGPT a mobility that triggers a web search, it starts by deconstructing your punctual into a group of its ain inheritance searches (fan-out queries), runs them successful parallel, and synthesizes an reply from immoderate comes back. Monitoring really fan-out queries alteration complete clip tells america a batch astir really OpenAI is tweaking the model’s hunt behaviour to effort to nutrient amended results. I deliberation watching this abstraction gives america a large hint astir what they were hoping to execute pinch each exemplary update.

It’s important to commencement by defining 2 words commonly thrown astir successful our space, without it ever being understood what the nuance is betwixt them. Retrieved intends a page ChatGPT fetched while moving its fan-out queries. Cited intends a page that made it into the visible reply arsenic a link. Citation and retrieval tin behave differently, and correct now they look to beryllium moving successful other directions: the number of “retrieved” URLs successful ChatGPT’s responses is growing, while independent measurements show the number of unsocial domains cited per consequence falling complete the aforesaid period. While much pages are being considered for the answer, less pages get cited and credited.

While reference immoderate study astir AI search, it’s worthy asking whether the article refers to retrieved URLs aliases cited URLs, and if it’s discussing citations, for which prompts those citations are appearing. In galore cases, the discrepancies I’ve seen betwixt fan-out studies travel down to 1 measuring retrieval and the different measuring citations.

To publication the existent fan-out queries (not conscionable the last answer), I’ve been utilizing a operation of a fewer tools: Peec AI and Profound some connection fan-out queries for tracked prompts, and the free Resoneo ChatGPT Chrome plugin and FanoutFox (shown below) some aboveground the queries the exemplary runs and the sources it pulls. These plugins make it easier to watch the exemplary “think” done its searches successful existent time.

Timing matters present too: it’s basal to see really and erstwhile ChatGPT releases caller models, and which models and tiers are astir commonly utilized by the mostly of its users. ChatGPT 5.6 ships successful much than 1 variant: Sol is the modular version, and the cheaper “Luna” version is what rolled retired astir the commencement of August arsenic the caller default exemplary forand Go users. That favoritism is important erstwhile you publication the studies below, because they aren’t each measuring the aforesaid exemplary aliases tier. And according to Olivier de Segonzac’s breakdown successful MCP, more than 90% of ChatGPT’s play users are connected the free plan. So immoderate the free default does erstwhile it searches, it’s now astir apt what the ample mostly of ChatGPT users get.

Part 1: What Industry Research Currently Shows

What I Saw In My Own Research

A fewer patterns stood retired to maine early on, earlier I went looking astatine anyone else’s data:

The amended models hunt more, and thin connected site: more. In my testing, ChatGPT 5.4 Thinking would occurrence 10-plus searches for a azygous prompt, including aggregate site: queries, while 5.3 Instant often did conscionable 2 to 3 fan-outs.
The type of site: hunt appears to dangle connected the query. For opinions and merchandise reviews, it leans heavy connected Reddit, including querying circumstantial subreddits. For YMYL-style questions, it appears to for illustration charismatic sources, often .gov domains aliases sometimes .org. In 1 batch of astir 20 prompts successful the ineligible space, every azygous 1 returned citations only from .gov domains. The beneath screenshot from the Resoneo plugin shows this process astatine work:

For specs and pricing, the exemplary goes to the brand. Searches for illustration site:sephora.com aliases site:costco.com showed up regularly. And if you inquire astir a merchandise aliases service, ChatGPT will often tally site: searches against the astir well-established brands successful that space, even erstwhile you didn’t sanction them successful your prompt. It besides often names circumstantial products and does site: searches for individual merchandise pages from the marque wrong the fan-out queries.

The nett effect, astatine slightest successful what I looked at, appears to beryllium much ChatGPT visibility for high-authority sites and trusted brands, and less citations for everyone else.

Below, I’ll item a fewer caller studies connected this taxable and what they found.

Useful Findings From SEO & AI Search Industry Experts On ChatGPT Fan-Outs

Several awesome folks successful our abstraction person been measuring this independently, pinch different devices and different postulation methods. David Konitzny astatine Peec AI ran the numbers the time ChatGPT 5.6 became the default: the stock of prompts pinch only a azygous fan-out query dropped from 94.0% to 43.5%, mean retrieved sources astir doubled from astir 12 to 24, prompts needing a 2nd fan-out loop went from astir 5% to 33.5%, and the site: usability went from appearing successful astir 0.3% of fan-outs to astir 23%. Basically, ChatGPT is becoming much precise and robust successful its searching process.

Chris Long astatine Nectiv, comparing astir 4,000 prompts connected 5.6 Sol against his ain 2025 baseline, recovered mean fan-out queries per punctual went from 2.17 to 7.61, the longest query concatenation went from 4 searches to 29, and “site:,” “official,” and “gov” each landed arsenic apical unigrams, pinch site: successful 64% of queries. There’s a large spread betwixt the site: hunt figures successful David’s and Chris’s articles, and it could beryllium explained by the fan-out query postulation method: Chris’s consultancy pulls fan-outs from OpenAI’s API, while respective of the different devices successful this abstraction extract them from the ChatGPT user interface. While the API is cleanable and repeatable, the UI method whitethorn beryllium person to what existent users really get (with definite limitations for illustration personalization, which nary instrumentality tin way effectively). It’s worthy checking which method a study utilized to understand why different studies whitethorn show discrepancies. In some studies, however, the stock of site: searches successful query fan-outs accrued substantially.

David besides recovered merchandise pages now dress up 16.39% of retrieved pages, moving up of listicles, which tracks pinch what I’d been seeing anecdotally: ChatGPT 5.6 appears to spell consecutive to brands and manufacturers for specs and pricing alternatively than routing everything done roundup articles and different listicles, which tin beryllium self-serving and prone to manipulation. The page types losing stock of retrievals (listicles, how-to guides, and comparison pages successful particular) also hap to beryllium the formats astir heavy spammed for GEO complete the past twelvemonth aliases two. In my talks passim this year, I’ve shared really these nonstop page types origin SEO and AI hunt problems.

Olivier de Segonzac and the Resoneo team, who person done immoderate of the astir elaborate reverse-engineering of the retrieval architecture retired there, recovered that the unsocial domains cited per consequence dropped from 19 to 15 aft the 5.3 update, confirming the shape from the different direction: retrieval counts grew while citation counts dropped. Ahrefs’ most-cited-domains data shows wherever the surviving citations onshore astir frequently: Reddit, Wikipedia, Forbes, Merriam-Webster, Consumer Reports, Healthline, and Walmart.

Suganthan Mohanadasan has been coming astatine it from the network-traffic side, and his findings are useful to understand: ChatGPT decides earlier it searches. Across 57 conversations and 3,554 retrieved pages, 21 of 27 first queries contained marque names the personification ne'er mentioned successful their prompt, crossed 11 of 13 merchandise categories. For example, erstwhile searching “the champion AI note-taking app,” the first fan-out query already contained the words “Granola,” “Notion AI,” “Otter,” “Fireflies,” “Fathom,” “Mem,” and “Limitless.”

Below is simply a snippet from Suganthan’s caller article, shared pinch permission:

Being the marque mentioned successful that first query is the existent goal: brands named successful the fan-out were cited 68.9% of the time, while pages that were only fetched were cited 2.1% of the time, and only 110 of the 3,554 retrieved pages, astir 3.1%, made it into an reply astatine all. Suganthan also explained ChatGPT’s routing logic wrong fan-out queries: facts way to charismatic pages, and opinions way to reviews and Reddit. So ChatGPT will look to different types of sites, and chopped pages wrong them, depending connected which accusation helps to champion reply the user’s question.

As pinch each studies, it’s adjuvant to publication his methodology alongside those numbers, which he states himself: a azygous account, based successful Dubai, sampled successful July 2026, weighted toward package and AI tools, pinch the brand-injection shape measured crossed 27 first queries. That’s a comparatively mini sample size, but it’s worthy citing because the guidance besides matches what the larger datasets show, and what I spot successful my ain testing.

The Consensus Between Recent ChatGPT Fan-Out Query Studies

Although the supra researchers utilized different tools, different models, and different postulation methods, their findings still statement up crossed 4 communal patterns:

ChatGPT is moving substantially much searches per prompt than it was a twelvemonth ago, including for free and cheaper tiers.
A meaningfully ample stock of those searches are now scoped pinch the site: operator.
The domains it scopes to skew high-authority: charismatic and shaper pages, established reappraisal platforms (G2, Clutch, Capterra, Consumer Reports, Wirecutter), Wikipedia, .gov and regulatory sources, awesome retailers, and Reddit.
Retrieval is going up while the number of unsocial domains cited is going down.

The skew is toward surfacing higher-quality, recognizable sites, and that inclination is continuing pinch newer ChatGPT models.

Part 2: My mentation On One Reason Why This Is Happening

I find each of this absorbing because it aligns intimately pinch 1 of the components of Google’s hunt ranking systems I’ve spent the astir clip studying: E-E-A-T (experience, expertise, authoritativeness, and trustworthiness). I deliberation OpenAI whitethorn beryllium utilizing fan-out queries arsenic a method of ensuring users get high-quality, trustworthy, charismatic information, while suppressing spammy and highly manipulated articles (such arsenic self-promotional listicles and different types of self-serving content).

I deliberation the site: usability is functioning (at slightest successful part) arsenic a spam filter.

For a while, ChatGPT would often retrieve and mention the nonstop benignant of spammy, self-serving, manipulative contented that we’ve already go rather acquainted pinch successful the SEO world. Evaluating the value of a random open-web page successful existent clip is genuinely difficult and expensive, and astatine ChatGPT’s measurement you’d person to do it hundreds of millions of times a day. Narrowing fan-out queries to .gov domains, established reappraisal platforms, large recognizable brands, and charismatic sources is presumably a overmuch cheaper measurement to get astir of the aforesaid outcome. Instead of judging whether an chartless page is trustworthy, the exemplary appears to beryllium sidestepping the problem wholly by only looking for accusation successful places it already trusts.

Google spent years building systems to reply “is this root charismatic capable for this query?” and what OpenAI appears to beryllium doing is simply a distilled type of this aforesaid process, relying heavy connected the site: usability and different keywords that power which sources get retrieved. For example: axenic facts spell to the brand’s ain domain. Health and ineligible questions spell to .gov sites. Opinions spell to Reddit and established reappraisal platforms. While Google uses analyzable ranking algorithms to aboveground high-quality, trustworthy, and charismatic pages, ChatGPT tin refine its fan-out queries to guarantee the hunt results propulsion from those trusted sources.

Adding “.gov” to definite queries (YMYL, perhaps?) is what I find intriguing, because it reminds maine of really Google elevates .gov (and different high-authority) sites successful its results for definite queries aliases during times of crisis. For example, during the Covid era, I shared a batch of research showing really Google elevated the charismatic FDA, CDC, and different high-authority and authorities sites successful the hunt results for health-related queries:

And the dependable emergence of the "real" authorities.

The .gov-only shape successful that batch of ineligible prompts (shown below) feels for illustration a large alteration connected ChatGPT’s part, arsenic it separates the astir “popular” aliases “optimized” pages connected the net from the charismatic authority sites. On almost immoderate user ineligible topic, the popular, top-ranking pages shown successful hunt travel from rule patient blogs, not authorities agencies. Combined pinch the page types losing retrieval share, that strikes maine arsenic a measurement for ChatGPT to trim done the sound and elevate charismatic accusation instead.

graphical personification interface, application

The refinement besides appears to beryllium moving down crossed the value tiers. The 5.4 Thinking exemplary was already doing a type of this: tons of fan-out queries, site: operators aimed astatine trusted domains, a clear penchant for charismatic and charismatic sources. With 5.6 becoming the free default, those hunt behaviors look to person besides extended to free tier users, not conscionable the users paying for the reasoning model.

The Use Of The Word “Official” In Fan-Out Queries

I recovered it very absorbing that the wave of the connection “official” successful fan-out queries appears to beryllium climbing, and it shows up arsenic a apical unigram successful Chris’s information alongside site: and gov. Similar investigation by Conductor, shown below, besides shows the summation successful site: searches and searches including “official.” ChatGPT seems to beryllium actively trying to find the charismatic root for a marque aliases product, which is precisely what you’d expect if the extremity is to debar citing unverifiable third-party pages.

This could coming a decent logic to revisit whether “Official” aliases “Official Site” belongs successful your homepage title tag aliases meta description, particularly for companies that stock a sanction pinch different brands aliases entities, aliases wherever the hunt results whitethorn different beryllium confusing. While this has already been a reasonably modular SEO proposal for a agelong time, I deliberation location are cases wherever making that connection definitive tin debar confusion. And fixed really overmuch ChatGPT now appears to beryllium looking for charismatic accusation successful the hunt results, it could perchance thief the exemplary reconcile which marque you really are, aliases which domain is officially yours.

The Risks Of Relying On Site: Searches

Heavy usage of site: searches successful fan-out queries tin activity good erstwhile the exemplary knows which sites to propulsion from. But articles by Malte Landwehr and Netcraft recovered that if location is disorder astir the correct domain for a fixed brand, ChatGPT appears to conjecture the domain, and sometimes it guesses incorrectly.

Malte Landwehr recovered examples wherever ChatGPT constructed site: queries against the incorrect marque domain, and successful immoderate cases, a domain that doesn’t beryllium to the institution astatine all. In 1 test, it many times searched site:census.com for the startup Census (not the authorities bureau), erstwhile the startup’s charismatic tract is getcensus.com, and census.com is parked and disposable to buy. He saw the aforesaid shape pointing astatine different parked domains for illustration lago.io, persona.id, lightfield.ai, and mesa.com.

A Netcraft study from 2025 besides recovered that astir a 3rd of marque login links generated by LLMs pointed to domains the marque didn’t own, and astir 29% pointed to unregistered, inactive, aliases parked domains, pinch smaller brands the astir exposed.

As Malte discussed successful his article, erstwhile the exemplary misidentifies the trusted tract for a lesser-known brand, personification could bargain that parked domain, big contented matching what ChatGPT is looking for, and softly provender it incorrect pricing aliases a clone support number. This could beryllium problematic for brands wherever the charismatic domain isn’t powerfully encoded yet, aliases wherever immoderate different logic prevents ChatGPT from identifying the correct domain name.

It’s worthy acknowledging that this narrowing could beryllium arsenic overmuch astir costs and latency arsenic it is astir quality. Checking a short database of known domains is cheaper and faster than evaluating the unfastened web astatine ChatGPT’s volume, and from the outside, that would look identical to a deliberate value filter. Malte’s uncovering besides shows location is still room to improve: a exemplary that was genuinely assessing whether a root is trustworthy and charismatic would beryllium little apt to constituent a hunt astatine a parked domain. So location could beryllium much than 1 logic why OpenAI is evolving really it constructs fan-out queries.

Site: Searches Appearing In Google Search Console

One site: query connected the tract beneath generated astir 197,000 impressions and precisely 1 click. That’s not the only example: when I looked astatine site: searches for respective awesome brands successful Google Search Console, I recovered thousands of impressions for site: queries pinch almost zero clicks, and the measurement of these queries is mostly expanding complete time. A click-through complaint that adjacent to zero is difficult to explicate pinch quality searchers, and points alternatively to bots (LLMs, search tools, scrapers, and truthful on). You tin spot akin site: hunt patterns successful Bing Webmaster Tools’ AI hunt query reporting.

This sewage maine thinking: ChatGPT is moving acold much site: searches than ever, and if it (or the partners it uses to scrape hunt engines for RAG) is still leaning connected a awesome hunt index, it seems plausible that astatine slightest immoderate of those searches would aboveground successful our query reporting. There’s nary charismatic “ChatGPT Search Console,” truthful this is speculation based connected looking astatine various accounts successful some Google and Bing’s hunt console tools, and Google has had its ain Search Console impression-related information issues muddying this nonstop benignant of data, truthful it’s difficult to get a conclusive reply arsenic to wherever these searches are coming from.

The applicable tip: Check some Google Search Console and Bing Webmaster Tools queries for site: searches and different repeated patterns. In Search Console, usage the Performance study and select Query by “Queries containing” site: (or a regex). In Bing Webmaster Tools, usage the Search Performance study and hunt the query array for site: strings.

While ChatGPT hunt has historically tally connected Bing (the OpenAI-Microsoft partnership), independent testing complete the past year suggests it’s besides pulling from Google’s index, perchance done third-party scrapers, truthful it genuinely isn’t clear which motor logged a fixed retrieval. This accusation is besides evolving each the time, particularly fixed Google’s suit against SerpApi, which was believed to beryllium the supplier of scraped Google information to OpenAI. ChatGPT besides appears to beryllium increasingly building and leveraging its ain index, which would springiness OpenAI much power complete freshness and sum without depending connected its biggest competitor.

Bing has made it easier to show AI hunt performance: successful February 2026, it launched an AI Performance Report wrong Webmaster Tools that separates AI citations from accepted hunt and surfaces “grounding queries,” which are the topics derived from hunt phrases Copilot generates internally to retrieve content. That was large news: a hunt motor choosing to study fan-out sub-queries arsenic their ain category, which tells you this full behaviour is existent capable that Microsoft is now building reporting astir it (while Google continues to hide it from us).

What I’d Actually Do About It

I spot a batch of this arsenic an denotation that SEO matters much than ever. Because ChatGPT appears to thin connected awesome hunt indices (directly aliases done RAG partners), ranking good successful hunt is still what feeds the fan-out, on pinch ensuring your marque has the correct accusation erstwhile its pages are straight searched for. Here are a fewer things I deliberation are worthy doing:

Build your marque to the constituent wherever it’s the 1 ChatGPT thinks of first. While this 1 seems beautiful obvious, this is wherever each the information points: you want your marque to beryllium synonymous pinch its category, and to beryllium people recommended capable to participate into the training information arsenic 1 of the astir trusted and well-known brands successful the space. Without bully branding, it’s apt that your marque won’t moreover participate the conversation. This is besides intolerable to triumph semipermanent via manipulative optimization tactics.
Put your cardinal facts successful plain, crawlable HTML connected your ain domain. Information specified arsenic pricing, specs, exemplary numbers, and support specifications needs to beryllium readable to the model. If ChatGPT is moving site:yourbrand.com looking for these, they request to beryllium included arsenic matter it tin really read, not locked wrong an image aliases client-side JavaScript-rendered content.
Nail down which domain is “official.” Consider “Official Site” (or similar) successful title tags aliases meta descriptions wherever it sounds people and helps users (don’t spam this), and make judge your existent domain is consistently reinforced crossed the web, peculiarly if you’re a smaller marque aliases stock a sanction pinch different entities. This is the azygous champion defense against the wrong-domain problem.
Don’t play whack-a-mole pinch trying to target individual fan-out queries. They’re long-tail, low-volume, and alteration betwixt prompts and users. Generating galore pages targeting the long-tail is simply a bully measurement to get caught up successful Google’s scaled contented abuse spam trap. It’s much useful to aggregate the queries the exemplary keeps moving and find the halfway topics it consistently prioritizes. This is what I judge Microsoft does successful the Bing Webmaster Tools AI hunt query report. I besides deliberation that aggregating fan-out queries and distilling them down into halfway keywords to usage for accepted SEO rank-tracking is still a bully method of approximating AI hunt performance: if you rank good for the topics mostly asked astir successful fan-out queries (ideally tracked crossed some Google and Bing), you person a higher chance of yet being portion of the AI response.
Check whether you are successful the query astatine all. Suganthan’s trial is simply a bully one: Run the prompts you attraction astir 5 abstracted times and look astatine whether your marque shows up successful ChatGPT’s first searches, not conscionable successful the answer. If it ne'er does, you apt person activity to do to build up the notoriety of your marque successful its niche, and that gets done pinch reviews, comparisons, integer PR, and media sum complete a agelong long of time, not pinch method fixes alone. If your marque does show up, the extremity is to person retrievals into citations and to power the AI response, and that is wherever on-page optimization tin count: authorities claims intelligibly and early connected the page, and guarantee important numbers and business specifications are retrievable for LLMs.
Take Reddit seriously. If the exemplary is pulling opinions from circumstantial subreddits, that’s portion of your brand’s visibility now, whether you for illustration the discussions aliases not. But retrieval and citation diverge here: Dan Petrovic recovered ChatGPT discards the Reddit pages it pulls astir 99% of the time. That truth stood retired to maine adjacent to Reddit topping Ahrefs’ most-cited-domains list, and some are existent astatine once: Reddit gets retrieved truthful relentlessly that moreover a ~1% endurance complaint produces much citations than almost anyone other successful absolute terms. It’s besides the aforesaid arena arsenic Suganthan’s 3.1% wide citation rate, seen from a azygous domain’s constituent of view. So dainty Reddit arsenic thing that shapes really the exemplary understands your class alternatively than arsenic a reliable citation path. And don’t usage immoderate of this arsenic an excuse to spam Reddit pinch artificial recommendations of your brand: Reddit precocious cracked down connected this nonstop type of GEO spam, utilizing its ain LLMs to emblem astir 25,000 spammy posts and comments per day, and playing pinch occurrence tin origin your relationship to beryllium banned.
Watch your site: impressions successful some Search Console and Bing Webmaster Tools. This is different illustration of bot activity affecting our Search Console data, connected apical of the caller discussions astir Google Search Console showing conversational prompts from AI Mode arsenic individual queries successful GSC. This is thing to see truthful the sound doesn’t distort your reporting, and it could potentially clasp immoderate accusation astir really ChatGPT (and possibly different LLMs) is retrieving accusation astir your brand, assuming that’s what that GSC information reveals.

I judge overmuch of this is simply a consequence to the clear flaws successful ChatGPT complete the past year, specified arsenic referencing biased and promotional content, for illustration listicles and comparisons that urge the aforesaid marque penning the article. These changes look to beryllium doing the opposite, narrowing difficult to .gov domains, large brands, Reddit, and charismatic sources, and extending that attack down from the premium models to the free default that astir of the world uses. While I don’t spot this arsenic anyplace adjacent arsenic blase arsenic Google’s mechanisms for achieving akin goals, it shows that OpenAI is innovating toward higher-quality results and moving connected solutions to the problem of manipulated contented appearing successful its responses.

More Resources:

How LLMs Interpret Content: How To Structure Information For AI Search
How ChatGPT Actually Picks Sources (I Read The Network Traffic, Not The Outputs)
ChatGPT Already Knows Who’s In The Running Before It Searches

This station was primitively published on Lily Ray NYC Substack.

Featured Image: CineVI/Shutterstock

Category SEO AI Search