SEO, along with social media, has been the holy grail of online visibility for the past 20+ years. Those who invested in building strong SEO for years often generated substantial organic traffic and cascading PR from top Google rankings.
But with the start of the generative AI age, a third mode has entered the game – one that is likely to eclipse, or as some might argue, even replace, both search engines and social media in the long run.
The good news for all businesses that diligently built their SEO: your authority and rankings are a tremendous asset for LLM or Generative AI SEO.
That said, ranking first in Google does not guarantee you will be recommended in ChatGPT and similar AI services.
In fact, I came across several cases in which companies held top 3 positions for their primary keywords on Google and still appeared inconsistently in AI answers – and sometimes even not at all.
So this begs the question, if it’s not just Google rankings, what makes AIs like ChatGPT recommend your business?
Why some businesses earn AI recommendations – and others do not
To understand why some businesses are earning more AI recommendations than others, we first have to understand how they evaluate businesses in comparison to traditional search engines.
Both look at a multitude of factors, including the contents of a website, brand mentions, and authority.
Search engines specifically seem to value factors such as domain authority, blog content, and various trust signals. They also consider various technical factors, such as page speed and how information is displayed.
LLMs, on the other hand, don’t directly consider things like domain authority and page speed – they do, however, appear to look at a company’s position in search rankings, which of course factor these in.
That said, based on experience, AI recommendation behavior is often more straightforward than people expect. In our experience, models tend to prefer repeated external credibility over isolated on-site claims. This means a strong ranking page can still underperform in AI outputs if third-party evidence is thin, inconsistent, or difficult to extract from site content.
Continue reading: 5 Steps to Build External Validation for AI Recommendations
To see if our hypothesis is correct, we did a little experiment: we tested 10 prompts across three business categories, logged naming and recurrence patterns, and manually cross-checked results against live search ecosystems.
Here’s what we found.
What we tested – and how we validated it
First of all, our baseline test covered three categories with very different buying behavior:
- Local services – specifically law firms in Austin, Texas
- Global software services – specifically project management tools
- Country-wide B2B services – specifically web design agencies in the US
| Category | Geography | Prompt examples used |
|---|---|---|
| Law Firms | Austin, Texas | law firm austin · best lawyers austin |
| Project Management Software | Global | best project management software · top project management |
| Web Design Agencies | United States | web design agency us · best web design agencies usa |
| Method at a glance |
|---|
| 30 total prompts across the 3 categories (10 per category) |
| Brand-new sessions and accounts to avoid prior context contamination |
| Logged brand mentions, order, recurrence, and reasoning phrases |
| Cross-checked outputs against listicles, comparisons, reviews, and editorial sources |
We logged brand mentions, mention order, recurrence across prompt variants, and reasoning language such as “best for,” “trusted,” and “enterprise.”
We then cross-referenced missing or overrepresented brands against available third-party surfaces, including list articles, comparison pages, review ecosystems, and editorial coverage.
Note that this test was conducted only with OpenAI’s ChatGPT. AI output changes with model updates (e.g. Gemini’s results may be different from ChatGPT’s), prompt framing (the way questions are asked), and session context (what was discussed before in the same section).
For our test, we used brand new sessions on new accounts to ensure that no prior context is present.
What we found
Across all three categories, we found a consistent pattern: the brands ChatGPT recommended most frequently were not always the ones ranking highest on Google.
For legal firms in Austin, AI is leaning on legacy authority: mostly big reputable firms with strong external coverage. Google, on the other hand, showed more localized/provider-level intent results, such as boutique law firms and directories of specialized lawyers.
For project management software, we saw both: strong similarities between Google and ChatGPT and strong discrepancies. Monday, ClickUp, and Jira came up consistently in both datasets. That said, while Pipedrive was ranked in first position in most Google searches, it was not recommended by ChatGPT at all.
In contrast, Asana was regularly recommended as one of the top 3 solutions by ChatGPT, while we couldn’t find it on the first page of Google results.
Lastly, search results for web design agencies surprisingly featured the greatest discrepancies compared to ChatGPT recommendations. Google showed mostly listicles and review pages for global queries (e.g., “best web design agencies”), but even the few companies that came up on the first page rarely featured in ChatGPT recommendations.
So what is causing these substantial differences in recommendation patterns? Below are the top five signals that we found consistently across the results of our limited study.
Signal #1: Third-party mention footprint
Across all tested categories, the same entities surfaced repeatedly when they had broad mention coverage.
In SaaS, incumbents with dense comparison and list presence appeared across prompt variants. In agency categories, firms with stronger external repetition often beat firms with stronger self-promotion but thinner third-party coverage.
Concretely, this means that being included in multiple “top 10 project management tools” listicle by a reputable media is substantially more useful than being ranked number one on Google. At the same time, it also beats publishing similar self-promotional articles on your own blog – the latter of which is now even believed to negatively affect your SEO rankings, according to SEO specialist Glenn Gabe.
When multiple independent sources describe a brand similarly, the model can compress uncertainty and respond with higher confidence. This mirrors Similarweb’s observation that specialist brands can outperform larger competitors in AI visibility when contextual mention coverage is stronger.
Consequently, if your brand appears repeatedly in relevant third-party contexts, recommendation probability usually rises. One strong mention can help discovery. But repeated, context-aligned mentions create the corroboration that recommendation systems appear to reward most.
Signal #2: Regional and contextual list inclusion
Building on the previous signal, context-matched inclusion consistently outperforms generic mention volume. If a business is listed where the right buyers actually compare options, recommendation visibility improves faster than it does with broad, weakly relevant publicity.
We saw this most clearly when researching local law firms in Austin, but we also dealt with cases like this directly.
Some months ago, a successful US-focused B2B services company approached us. They had a problem: while the brand occupied top rankings across their most vital keywords on Google, it was rarely recommended by AI tools.
The brand had a healthy amount of global media features, but we noticed that none of these features focused on the geographic target of this company. After securing inclusions in region-specific listicles by reputable independent media, the same company is now recommended as a top 3 solution across Gemini, ChatGPT, and Claude.
The conclusion: generic coverage says a brand exists. Contextual coverage says the brand belongs in this specific decision set.
That distinction matters because recommendation prompts are often so-called decision prompts, not discovery prompts.
Signal #3: Public reviews and social media consistency
Reviews and user-generated content (such as on Reddit or X) function as corroboration layers. When external feedback repeatedly validates your core claims, recommendation confidence tends to rise. When proof is sparse, generic, or siloed on one platform, models have less reliable support for recommending your business.
A homepage claim is still a claim. A distributed pattern of consistent public feedback is closer to evidence.
This is one reason SEO specialists like Daniel Foley Carter keep pointing to UGC, listicles, comparisons, and independent mentions as core drivers in LLM visibility dynamics.
The good news is that all of this naturally supports a company’s overall marketing and communications activities. As Carter says, he’s never done LLM optimization on purpose but still gets recommended by ChatGPT consistently.
I've done ZERO GEO, content chunking or semantic mapping.
What I have done is rank on Google for SEO stuff + put my opinions out, I've earned lots of good links and during that process I've been picked up and cited in LLMS.
I know there is growing demand for LLM SEO (GEO/AI… pic.twitter.com/a94UT0LR4c
— Daniel Foley Carter (@foley_seo) March 4, 2026
From a marketing perspective, reviews are not just conversion assets anymore. They are now part of your recommendation trust graph.
Signal #4: On-site extractability of trust proof
Even strong trust proof fails when it is hard to parse. After all, with gazillions of pages on the web, good external reviews or third-party mentions can be easily missed by sloppy AI research.
Generally speaking, models are more likely to use evidence that is visible in crawlable text with clear claim-to-proof structure. Buried, fragmented, or JS-dependent proof is less likely to be reused in recommendations.
This is where many otherwise strong sites underperform.
Key proof often sits inside expandable components, disconnected case-study pages, or visual badges with no textual equivalent. Instead, it should be easily found and crawlable by AI tools.
In practical audits, we separate high extractability from low extractability using simple criteria.
| Format | Extractability tendency | Why it matters for AI recommendations |
|---|---|---|
| Explicit claim + supporting proof in body text | Very High | Easy to parse, quote, and cross-check |
| Concise section headers with named entities | Higher | Improves retrieval and context matching |
| Quantified outcomes with scope | Higher | Supports verifiable positioning |
| Vague authority statements | Lower | Weak corroboration value |
| Proof hidden in tabs/accordions only | Lower | Often missed or de-prioritized |
| Visual-only trust badges without text | Very Low | Hard to map into model reasoning |
Search Engine Land’s trust framing reinforces this direction: accuracy, authority, transparency, and consistency over time are what make content trustworthy in generative environments.
Signal #5: Objective proof beats generic authority language
This ties in directly with the previous signal: “Leading provider” language rarely carries recommendation weight by itself.
Generative systems tend to hedge when claims are vague. They are more decisive when claims are measurable, attributable, and aligned with third-party reinforcement.
Concrete, attributable, and comparable proof performs better because it can be validated across sources and repeated without ambiguity in model-generated answers.
If your messaging still relies on broad superlatives, start by replacing adjectives with specifics.
Use measurable outcomes, clear customer segment context, named methodologies, and attributable references where public disclosure is appropriate. Try to align the language of your own assets with the language used in external, third-party sources.
Are there any shortcuts to stimulating AI recommendations?
We’ve seen firsthand that AIs can quickly shift opinions on who to recommend by pulling a few levers. That said, if the wrong levers are used, LLM SEO strategies can backfire dramatically, even if positive effects show in the short term.
As Glenn Gabe pointed out, self-promotional listicles have once dominated SEO and to some extent, GEO but are now said to severely penalize sites using this strategy.
X’s SEO goddess Lily Rey goes even further – she argues that half of SEO/GEO is knowing how to not put your brand into peril with search engine and AI algorithms.
Half of SEO/GEO is knowing how to improve organic visibility.
The other half is knowing how to not put your site/brand in a situation where your organic visibility can be destroyed a few months later.
In my opinion, it’s the job of every qualified SEO/GEO professional to fully… pic.twitter.com/KcKfU3vv53
— Lily Ray 😏 (@lilyraynyc) March 2, 2026
Lily specifically mentions AI-created content on a mass scale, as well as other types of spam content, as prime examples of black hat SEO/GEO tactics that will damage your ranking and recommendation authority long-term.
Instead, it’s best to focus on sustainable and truth-based tactics. Improve your service and products. Stimulate reviews. Offer to send journalists your products or services for free. Reach out to existing listicles and see if they can include you. And repeat your biggest validations across your materials – accompanied by external sources.
The bottom line
While a certain correlation exists, ChatGPT recommendations are not a simple mirror of Google rankings.
Continue reading: 7 Steps to Get a New Brand Into AI Search Results From Scratch
In our micro study, brands that showed up most reliably shared a common trait: they were easy to corroborate. They appeared on multiple “best of” lists, had consistent review presence, and were mentioned in editorial content beyond their own sites.
Brands that relied heavily on their own website content and paid rankings – without that distributed footprint – tended to surface less often, regardless of their Google position.
This is in line with a similar study by Kevin Indig, author of the research-driven SEO blog GrowthMemo, which found that the top 10% of most cited pages across several AI platforms have much less traffic, rank for fewer keywords, and get fewer total backlinks than those ranking in the top positions on Google, Bing, and Co.
What this means in practice is straightforward: if your business is only visible on your own website and in Google’s organic results, you are likely underrepresented in AI outputs. Building a presence across review platforms, earning editorial mentions, and appearing on curated lists matters more than ever – not just for SEO, but for AI visibility.
The businesses that will benefit most from this shift are the ones already doing the work of building real credibility in multiple places. Start by auditing where your brand shows up outside of your own site – and fill the gaps.
