Bing, not Google, shapes which brands ChatGPT recommends

On this case examine, we went deep as an alternative of broad. We centered on one query: why wasn’t a model current in a single ChatGPT immediate throughout ~70 iterations?

We selected one immediate: “What are one of the best inns in New York Metropolis?” 

We analyzed mentions, citations, fanouts, and SERPs in Google and Bing. We additionally deliberate to research GPT reminiscence, nevertheless it made no discernible distinction to mentions, citations, or fanouts.

What we did and what we discovered

We selected NYC inns as a result of it’s a crowded, mature market with juggernauts and up-and-comers. We additionally haven’t any connection to the NYC luxurious lodge area — we deliberately picked an space the place we may keep goal and study from scratch.

After operating the immediate “what are one of the best inns in New York Metropolis” 68 occasions, we recognized which inns appeared most constantly and which have been almost invisible.

We selected the Baccarat Lodge as our “shopper” as a result of it appeared solely as soon as (1.5% of the time), regardless of sturdy opinions and clear alignment with the prompt’s intent. We wished to know why — and whether or not it may change that.

Key findings:

• You’ll be able to dominate question fanouts on Google SERPs and nonetheless underperform in ChatGPT model mentions.
• Bing issues most. Rating in Bing articles for fanouts aligns extra straight with ChatGPT mentions — not simply citations.
• In verticals dominated by third-party content material, you face complicated digital PR paths to extend visibility.

Observe: A full methodology breakdown seems within the appendix.

Mentions of the Baccarat vs. the Fifth Avenue Lodge present simply how broad the disparity in ChatGPT visibility could be

The Baccarat Lodge appeared as soon as in 68 trials (1.5%).

Prime performers have been massive luxurious inns just like the 4 Seasons Lodge New York Downtown.

ChatGPT additionally recognized boutique inns as a subcategory, producing a secondary checklist in its solutions. Boutique inns just like the Baccarat are sometimes smaller and never a part of massive chains.

Inside this boutique subcategory, the Baccarat nonetheless underperformed. The Fifth Avenue Lodge, the top-performing boutique property, appeared 13 occasions, cited 20% of the time, versus the Baccarat’s 1.5%.

Status can’t clarify visibility disparities

We first checked whether or not something within the lodge’s historical past or fame may clarify the hole. Because the chart beneath reveals, nothing important did:

General, the Baccarat has been round longer and has extra opinions. On high quality, the Fifth Avenue Lodge has no edge in Google opinions and solely a slight edge in Expedia opinions. The one space the place the Baccarat lags is worth — however that’s unlikely the difficulty when The Ritz-Carlton, a constant non-boutique winner, is listed at $1,100.

Additional reinforcing the Fifth Avenue’s underdog standing: one among its most outstanding Google outcomes (rank 2) was a Wikipedia web page for a special Fifth Avenue Lodge that closed in 1908, creating potential entity confusion much like the two Danny Goodwins.

If the Fifth Avenue Lodge had been the one lacking, it might recommend a much less established model with entity confusion. However the reverse occurred — it prevailed in ChatGPT.

So what was the issue for the Baccarat Lodge?

Profitable Google SERPs for question fanouts doesn’t assist, however successful Bing SERPs does 

When ChatGPT performs an online search, it sends a sequence of queries you may extract through Chrome DevTools. On this case examine, examples included:

• [Best hotels in new york city]
• [Top rated luxury hotels in new york city recommendations]
• [Best hotels in nyc top luxury and boutique hotels new york]
• [Best luxury and boutique hotels in new york city recommendations reviews]
• [Best hotels in new york city nyc top hotels]
• [Top hotels in nyc luxury boutique best places to stay new york city]

In complete, we extracted 25 distinctive question fanouts.

What we noticed within the Google SERPs

If we solely regarded on the articles dominating fanout SERPs in Google, we’d anticipate the Baccarat to narrowly outperform the Fifth Avenue in ChatGPT. That didn’t occur.

Within the desk beneath, the Baccarat “wins” three of the highest 10 most ceaselessly showing pages, whereas the Fifth Avenue Lodge “wins” two. The opposite 5 function neither. A “win” means one of many following:

• Showing when the opposite doesn’t.
• Showing increased on the web page.
• Having extra constructive sentiment.

The info:

What we noticed within the Bing SERPs

In contrast, trying solely on the articles dominating fanout SERPs in Bing, we’d anticipate the Fifth Avenue to outperform the Baccarat in ChatGPT — and it did.

Within the desk beneath, the Fifth Avenue “wins” 5 of the eight most ceaselessly showing URLs.

Observe: The desk consists of two fewer URLs as a result of Bing SERPs have been barely much less numerous for these fanouts.

The info:

The connection between Bing visibility and model mentions

Bing rank strongly predicts ChatGPT citations — 87% align with Bing’s prime outcomes, Seer Interactive found. Our case examine helps this and extends it.

We examined the connection between fanouts (Seer centered on prompts) and model mentions.

Instance point out: “For a luxurious boutique really feel: listings like The Fifth Avenue Lodge or Crosby Avenue Lodge constantly make ‘prime NYC’ lists from journey editors.”

Mentions are sometimes extra useful than citations. Most individuals received’t comply with citations however will bear in mind the highest advice.

There’s ongoing debate about whether or not fanouts form ChatGPT’s solutions and mentions, or just assist solutions generated from coaching information. For instance, Leigh McKenzie argued on LinkedIn:

• “The citations you see on the backside? These are surfaced after the reply is generated, not earlier than. It’s post-hoc rationalization. The mannequin didn’t select your model as a result of it discovered your URL. It generated a solution based mostly on what it already is aware of, then pointed to sources that assist it.”

In contrast, our information aligns with Beehiiv’s research, which suggests citations do form mentions.

Coaching information doesn’t seem like the difficulty for the Baccarat. In comparison with the Fifth Avenue, it’s older, has extra opinions, and holds equally excessive rankings throughout main platforms. What it lacks is robust presence in Bing outcomes for fanouts and citations, which seems to result in fewer mentions.

A easy stream may appear to be this:

• Model ranks in Bing → ChatGPT fanouts pull in Bing pages → ChatGPT synthesizes coaching and Bing information to generate mentions

Coda: A story of two Forbes articles, or why the main points matter

On this vertical, third events like Forbes and Condé Nast management the area. Visibility is determined by who mentions you, so that you want a powerful outreach technique — not simply updates to your individual content material.

Our information reveals that “concentrating on Forbes” isn’t particular sufficient.

The highest end result surfaced in each Bing and ChatGPT was the identical Forbes article. In Google, probably the most frequent fanout end result was additionally a Forbes article — however a special one.

As we’ve seen, stepping into Google’s Forbes article possible wouldn’t present a significant increase. The Baccarat “received” in that piece.

Moving into Bing’s Forbes article, the place the Baccarat wasn’t talked about, may make all of the distinction. This requires a extremely surgical strategy grounded in Bing information.

Generalities received’t work; element reigns supreme.

Appendix: Methodology

Mannequin: We prompted GPT-5.2 On the spot and manually extracted outcomes. We didn’t use APIs inside ChatGPT.

Variety of iterations: We ran the identical immediate 68 occasions.

Immediate: “What are one of the best inns in New York Metropolis?”

Settings: We examined three reminiscence states:

• Saved recollections off
• Saved recollections on, utilizing unrelated actual person recollections
• Saved recollections on, with one reminiscence about needing gluten-free journey lodging

For all trials, we turned off “reference chat historical past” to keep away from interference throughout iterations.

We anticipated variations based mostly on reminiscence settings however discovered none, so we handled all trials as a single dataset.

What we extracted:

• All question fanouts.
• Full ChatGPT textual content output.
• Citations.
• Google SERPs for all fanouts.
• Bing SERPs for all fanouts.