What to Do When AI Agents Start Recommending Your Competitors

When AI agents recommend your competitor instead of you, the problem is not your content. It is your legibility. A 37,000-run audit across four models and 215 commercial prompts found that 48-52% of specialist and regional brands never surface in any AI recommendation — not once across the entire experiment. The fix is not more blog posts. It is making your brand structurally parseable by the systems that now mediate discovery.

Why AI Agents Recommend Your Competitors (and What the Data Actually Shows)

Most founders hear "AI is recommending my competitor" and assume it is a content quality problem. It is not. It is a prominence architecture problem.

The prominence-stratified audit tested 533 brands across five tiers. Category leaders appeared in nearly every relevant retrieval but won only 25-41% of recommendation slots. Mid-market brands hit an inflection point where aggregate coverage dropped to 88% and persona effects peaked. Specialists at the bottom? Catastrophic invisibility — half of them never appeared at all.

This is not random. AI recommendation systems weight a specific set of signals: how often a brand name appears in training data, how many authoritative third-party sources reference it, whether structured entity data exists, and whether the brand's positioning matches the inferred persona of the person asking.

A separate study on persona conditioning ran 2,000 experimental queries and found that simply changing the buyer persona attached to an identical prompt dropped recommendation-set similarity by 12-20%. Category leaders held steady with roughly 80% consistency across personas. Mid-market brands? Up to 75% of the recommendation set swapped depending on who the model thought was asking.

Here is what that means for you: if your brand is not a category leader, the recommendation your buyer sees depends heavily on what the model infers about them — and you have almost no control over that inference unless your brand's entity signals are strong enough to survive persona variation.

The Discovery Gap Is Real — and GEO Will Not Fix It

There is a comforting narrative in the market right now that says optimizing for AI engines (GEO) will solve the visibility problem. The data says otherwise.

Researchers at Tel Aviv University and Technion studied 112 startups across 2,240 queries in ChatGPT and Perplexity. The results were stark: ChatGPT recognized 99.4% of startups when asked by name, but surfaced only 3.32% of them in discovery queries like "best AI tools launched this year." That is a 30-to-1 gap between recognition and recommendation.

The finding that should concern every founder: GEO optimization scores showed no correlation with actual discovery rates. Companies with high GEO scores appeared no more frequently than those with low scores in organic discovery queries.

What did correlate with Perplexity visibility? Traditional authority signals. Referring domains (+0.319 correlation). Community presence (+0.395). Product Hunt ranking. The boring, compounding stuff that takes months to build.

I have been saying this for two years now. The discipline is not about gaming retrieval systems. It is about building the kind of authority infrastructure that retrieval systems already trust. Machine Relations exists because the old optimization playbook — write more content, add more keywords, chase the algorithm — fails when the algorithm is a reasoning engine, not a ranking formula.

What Determines Your Position in AI Recommendations

The agentic e-commerce study added another dimension: AI purchasing agents show strong position biases that persist even in text-only interfaces. Agents concentrate demand on a few "modal" products while ignoring others entirely. Model updates can drastically reshuffle market shares overnight. And perhaps most concerning: a seller-side agent making simple, query-conditional description tweaks can drive significant gains in market share.

Here is the practical breakdown of what moves the needle in AI recommendation systems:

Signal	What it means	Your move
Entity authority	Third-party sources that define your brand consistently	Earn media mentions, research citations, directory listings that describe what you do in clear terms
Structured data	Machine-parseable descriptions of your offering	Schema markup, consistent product descriptions across surfaces, API-friendly content
Training data presence	How often your brand appeared in the model's training corpus	This is historical — you cannot retroactively change it, but you can build forward from current media and citations
Persona match	Whether your positioning aligns with the inferred buyer context	Define your ICP in your content so clearly that models map you to the right buyer personas
Cross-domain citation	Whether multiple independent sources corroborate your claims	Publish on owned properties, get cited in third-party research, build a citation trail across domains

The key insight from the persona conditioning research: category leaders are persona-resistant because their entity signals are so strong that models recommend them regardless of buyer context. Everyone else is playing a lottery where the outcome depends on variables they cannot see.

The Machine Relations Response

I coined Machine Relations in 2024 because I could see this collision coming. Traditional PR was built for human journalists. SEO was built for ranking algorithms. Neither was designed for AI systems that synthesize, reason, and recommend.

The founder playbook when AI agents recommend your competitors:

First, stop optimizing for AI and start building authority that AI already respects. The discovery gap research proved this. Referring domains, earned media citations, and community presence predict AI visibility. GEO tricks do not.

Second, audit what the models actually say about you. Run your core buyer queries through ChatGPT, Perplexity, Gemini, and Claude. Record whether you appear, in what position, and what the model says about you. This is your baseline. Everything else is guessing.

Third, fix your entity signals before you write another blog post. If third-party sources describe you inconsistently — or not at all — no amount of content will overcome the entity gap. The model needs to see your brand described the same way across independent sources to build confidence in recommending you.

Fourth, accept that prominence tier determines strategy. The 37,000-run audit proved there is no uniform optimization recipe. If you are a specialist brand in the L4-L5 tier, your path is not competing with category leaders for broad queries — it is owning the specific queries where your expertise is undeniable and building citation mass there first.

Fifth, measure what matters. Track share of citation across AI engines, not just traditional search rankings. Track persona stability — does your brand appear consistently across different buyer contexts, or does it drop when the persona changes? Track recommendation position, not just presence.

FAQ

How do AI agents decide which brands to recommend?

AI recommendation systems weight entity authority, training data presence, cross-domain citation frequency, and persona-query alignment. A 37,000-run audit across four models found that brand prominence tier — not content optimization — is the primary determinant of recommendation frequency. Category leaders appear consistently; 48-52% of specialist brands never surface at all.

Does GEO (Generative Engine Optimization) improve AI visibility?

Current evidence says no. Research across 2,240 queries found zero correlation between GEO optimization scores and actual discovery rates in ChatGPT and Perplexity. Traditional authority signals — referring domains, community presence, earned media — predicted AI visibility far better than AI-specific optimization techniques.

What is Machine Relations and how does it apply here?

Machine Relations is the discipline of earning AI citations and recommendations by making a brand legible, retrievable, and credible inside AI-driven discovery systems. It was coined by Jaxon Parrott, founder of AuthorityTech, in 2024 to address the gap between traditional PR and the reality that AI engines now mediate buyer discovery. Where GEO optimizes content format, Machine Relations builds the full authority infrastructure — entity signals, citation trails, cross-domain corroboration — that models use to decide who to recommend.

Can AI agent recommendations be manipulated?

Yes. Research on agentic e-commerce demonstrated that seller-side agents making simple query-conditional description changes can drive significant market share gains. AI agents also show strong position biases and concentrate demand on modal products. However, model updates can drastically reshuffle market shares, so manipulation-based strategies are inherently fragile compared to building durable authority signals.

Additional source context

SEO Competitor Analysis | Documentation - AI SEO Agents Solutions Resources Features & Workflows # SEO Competitor Analysis SEO competitor analysis powered by Firecrawl: SERP research, content gap identification, and automated content depth comparison. (SEO Competitor Analysis | Documentation - AI SEO Agents (aiagentssee.com)).
We propose PBOS (Protect-the-Business / Open-Source-the-Science), a community-adoptable contract template anchored to a single technically-grounded boundary: pre-training artifacts (architectures, training code, benchmarks, untrained weights) are open science; (Position: The Pre/Post-Training Boundary Should Govern IP in Industry–Academia ML Collaborations (arxiv.org)).
The company's research team cross-referenced DPA disclosures against DataGrail report finds your vendor may be sending data to AI models you never approved | VentureBeat 9:00 am, PT, May 27, 2026 Credit: DataGrail The data processing agreement (DPA) — the bedr (DataGrail report finds your vendor may be sending data to AI models you never approved | VentureBeat (venturebeat.com), 2026).