The 2026 Agent Image
Visibility Index

1How to Read This Study

This is not a critique of any agent, team, or brokerage. The portfolios here belong to accomplished real estate professionals, many of them top producers with beautiful, well-built websites.

It is also not, at heart, a critique of a website platform. Agent Image builds polished, custom-designed sites, and our data confirms the container is competently made. The question is narrower, and modern:

Is your digital presence structured so today's AI answer engines can understand you, trust you, and recommend you to a buyer who does not already know your name?

We measure structure, not style, and we focused on a single platform on purpose. One month earlier we ran the identical study on 100+ sites built by Luxury Presence, the platform of choice for elite luxury agents, and found near-total structural invisibility to AI. The fair objection was: maybe that is one vendor's problem, so look at another. So we held a different platform constant: 100 sites, one premium vendor, Agent Image. The pattern does not just repeat here. It is worse. It is not just Agent Image. Agent Image is simply where we pointed the microscope this time.

2Why This Matters

A website audit sounds like a technical exercise. It is now an existential one, because the buyer has moved to AI and most agents have not.

The buyer has already moved

• 82% of Americans active in buying or selling now use AI for housing-market information (among them, ChatGPT is the most-used platform at 67%, Gemini at 51%), per an independent Realtor.com survey of 1,000 U.S. adults in August 2025.¹
• Buyer research is increasingly starting in AI rather than Google. Industry benchmarks put AI-first buyer searches at roughly 61%, and use of AI as the primary tool for researching an agent has climbed from 17% to about 67% in 18 months.²

It is fair to note that these engines are still a tiny slice of search today. Organic search sends well over a hundred times more traffic than all AI engines combined. But that is the wrong lens, because the trend is what matters. AI referral traffic has grown roughly sixteenfold in two years, and the leading engines are climbing 200 to 320 percent year over year.⁴ The shift is not a question of whether, but of how fast, and the agents who engineer their visibility first are the ones the machines will already know by the time it is mainstream.

The window is still open

That same benchmark found that roughly 71% of U.S. metros have no agent with dominant AI visibility yet. For an agent willing to engineer it now, the AI shortlist for an entire market is still unclaimed. The window closes as competitors build entity authority first.²

Why agents do not see the problem themselves

Most agents test their AI visibility by searching for themselves, and conclude they are fine. They are not seeing what a buyer sees. Since April 2025, ChatGPT personalizes web-search results using memory and prior conversations, so an agent who has discussed their own business with ChatGPT receives an inflated, self-referential result.³ A neutral buyer, on a clean device with no history, sees something entirely different, which for the great majority of agents is nothing at all. This personalization gap is exactly why a real visibility audit must be run on a neutral session, not on the agent's own screen.

Sources: 1. Realtor.com, "AI and Housing Survey," August 2025 (realtor.com/research/ai-and-housing-survey-2025). 2. FlyDragon 2026 AI Search Benchmark, 8.2 million queries across 192 metros (newswire.com). 3. TechCrunch, "ChatGPT will now use its memory to personalize web searches," April 18, 2025. 4. SE Ranking, "AI Traffic Research Study," June 2026 (analysis of 101,574 websites, January 2025 to April 2026).

3Two Platforms, One Verdict

This report is a deep dive on a single platform, Agent Image. It is the second such deep dive in two months, and reading it alongside the first is what turns an observation into a controlled finding.

June 2026: the backdrop, the elite's preferred vendor

One month before this study, we ran the identical three-part audit on 100+ sites built by Luxury Presence, the website platform chosen by top-producing luxury agents and luxury brokerages. We selected the platform's elite, names with billions in collective career sales, and ran 2,040 live AI answers against them. The finding was stark: 166,742 technical errors, and 55% of those elite portfolios invisible to a non-branded AI query. A beautiful, fast, premium container, and more than half of it dark to a buyer who did not already know the agent's name.

That raised the obvious question. If the most decorated agents in the country are invisible on their premium platform, is any other premium platform actually different, or is this structural to how these site builders ship by default?

July 2026: this report, a second premium platform

So we held a different vendor constant and went deep on Agent Image, another established premium builder with a long roster of top-producing agents, teams, and boutiques, several featured by the platform itself as showcase work. We audited 100 of them: a full technical crawl of each site, 2,000 live AI answers across four engines, and an organic-authority read of every domain. Same methodology, same rulers, different platform.

Two premium platforms, two deep dives, one verdict: the website is not the problem, and the website is not the solution. The engineered structure underneath is. Change the vendor and the failure does not go away, it deepens.

4Executive Summary

This is the July 2026 deep dive in brief: 100 Agent Image portfolios across roughly 40 U.S. and international markets, 2,000 live AI answers across ChatGPT, Gemini, Perplexity, and Grok. A month after the Luxury Presence study, on a second premium platform, the pattern is unchanged, only more severe.

These portfolios are not suffering from poor design. They are suffering from structural invisibility, and most function as a glorified business card: findable when a buyer already knows your name, absent the moment discovery depends on the machine knowing who you are.

Recognized by name, invisible to a buyer

Asked about by name, these portfolios surface 33.0% of the time. Asked the question a real buyer asks, "best agent in [city]," they surface just 6.8% of the time. That gap is the business card.

The two biggest engines are blind to them

Of 100 portfolios, only 7 ever surfaced on ChatGPT non-branded, and 4 on Gemini. Nearly all faint signal lives on Perplexity, which scrapes the live web rather than answering from trained memory. These sites have given the model a name to recognize, never an entityiA specific real-world thing an AI recognizes, like a named person or business, not just words on a page. AI recommends entities it actually knows. to recommend.

A shared signature

100 different agents, teams, and boutiques, with different reputations and different budgets, yet the failure signature is nearly identical: the same image, link, and metadata gaps in the same proportions, site after site. When the only thing held constant is the platform and the failure does not vary, the platform's default structure is the explanation. A beautiful container is not an engineered asset.

5The Glorified Business Card

We tested one hypothesis: that a site on this platform behaves like a business card. It works when someone has your name, and returns nothing when a buyer asks a generic question. Across all 100 portfolios and 2,000 AI answers, the data confirms it, more sharply than on the first platform.

A buyer relocating to one of these markets opens ChatGPT and asks, "Who are the best agents in [city]?" For nearly three in four of these portfolios, the answer never includes them. The AI returns the national brokerages and portals, Compass, Sotheby's, Coldwell Banker, Zillow, which own the generic answer because they have the entity structure these individual sites lack.

Agent Image gives the AI a name to recognize. It does not give it an entity to recommend. The platform produces a card that works when someone already knows you, and vanishes the instant discovery depends on the machine knowing who you are.

6The Single-Engine Illusion

Discovery no longer happens on one surface. A buyer may ask ChatGPT, Gemini, Perplexity, or Grok, and the answers differ sharply.

The two most-used consumer engines recommend virtually none of these agents non-branded. Whatever visibility exists is borrowed, in real time, from Perplexity's crawl. It is not owned in the model's memory of who matters in a market.

A name the model has not learned is a name the model cannot recommend. Live-web scraping papers over the gap on one engine. It does not close it.

7The Three States of Digital Existence

Sorted by live organic performance, the 100 portfolios fall into three states.

The roster is authority-poor: median Domain RatingiDomain Rating (DR): a 0 to 100 score for how trusted a website is, based mostly on how many other good websites link to it. Higher is better, like a credit score for a website. (DR) 12, and 32 of 100 sites draw zero measurable organic traffic at all. More than half the cohort is a Ghost Town, organically invisible before AI even enters the picture. This is the sharpest contrast with the first platform, where Ghost Towns were 10% of the roster. Here they are the majority.

The Traffic-Leader Paradox

Organic traffic and AI recommendation are now separate currencies. This platform earns the first for a handful of sites while forfeiting the second, and the traffic it does earn is mostly branded: agents rank for their own name, not for the buyer's question.

8Strategic Conclusions

1 · The non-branded epidemic is not just real, it is deepening

72% of these portfolios are invisible to a non-branded AI query, and 22% are dark even by name. On the first premium platform we measured, those figures were 55% and 18%. Same failure, larger. To a buyer asking an AI who to call, nearly three in four of these agents do not exist.

2 · Fame is not enough, the recognition trap

A nationally televised luxury brokerage, one of the most recognizable names in the business, draws tens of thousands of monthly visits and returns zero non-branded AI visibility. On some engines the model recommends the agent under a different web address than the one the platform built, so the site itself scores nothing. You cannot PR your way onto the AI shortlist, and a famous name pointed at the wrong domain is worse than no name at all.

3 · The platform is the common denominator

100 different operators, one identical failure signature. The same handful of image, metadata, and markup gaps fire on essentially every page of essentially every site, regardless of the agent behind it. When the controlled variable is the platform and the outcome does not vary, the platform's default structure is the explanation.

4 · It is not just this platform

Our June 2026 study found the same pattern on a different premium vendor, and our December 2025 industry sweep found it across many. No container, however custom, ships the engineered authority an AI requires. That layer has to be built.

5 · Nothing has changed, and that is the point

One month after we first measured this on a premium platform, on a second premium platform, the numbers are no better and in most columns they are worse. Early movers who build entity authority now become the default AI recommendation in their markets. Everyone else is accepting the business card.

Appendix A · Methodology & Data Integrity

All findings draw exclusively from primary data we collected, with no external benchmarks or speculative models. The external figures in "Why This Matters" are cited to their sources and are not part of our own dataset.

Site selection. The 100 portfolios were chosen deliberately, not at random: they are established agents, teams, and boutiques on Agent Image, including several of the platform's own showcase clients. This is a resourced, professional slice of the market by design. It also makes the finding conservative. If well-established agents on the platform are invisible to AI, the agents below them are not faring better.

• Technical audits, a full crawl of each domain with every signal categorized. 84 of 100 sites crawled successfully; 16 could not be crawled by our tooling (a mix of bot-protection, SSL, and server errors) and are excluded from technical totals while retained in the AI set.
• AI visibility, five prompts per site across four engines, 20 answers per site, 2,000 total. Four prompts non-branded ("best agent in [city]"), one branded ("tell me about [agent]"). A site counted as visible only when an engine recommended its exact registered domain, not when a URL merely appeared as a scraped citation.
• Organic authority, Domain Rating, monthly organic traffic, and keyword footprint for every domain.

Disclosures. Microsoft Copilot and Google AI Mode are not exposed by current tooling and are excluded. One site returned empty results from a single engine on every prompt and was scored zero for that engine; all other data is complete. We make no claim about AI-crawler access policies. An automated flag suggested some sites block AI crawlers, but raw robots.txt inspection showed those were benign utility-path rules, not real blocking, so we do not report a crawler-wall finding. The exact-domain visibility ruler is deliberately strict and undercounts agents recommended under a different web address; see Appendix F.

Appendix B · Full Error Taxonomy

118,421 signals across 84 crawled sites, about 1,410 per site, about 15 per page. The top three categories account for 50.7% of everything, and the classic "big three" of images, links, and social metadata account for 46.3%.

The failure is not evenly spread. It concentrates in the categories that control how a machine reads identity: social metadata, markup hygiene, schema, and images. That is the signature of a site built to look right to a human and never engineered to be legible to an AI.

Appendix C · Branded vs. Non-Branded & Engine Matrix

Non-branded reach by engine, with a denominator of 400 prompts per engine:

The exceptions prove the ceiling is reachable. A Los Angeles luxury agent with genuine national press scored 9 of 16 non-branded; a Beverly Hills brokerage with deep media citations scored 9 of 16; a small-market specialist dominating a single resort town scored 8 of 16. In every case the lever was off-site authority, press, and entity definition, not anything the platform ships by default.

Appendix D · Organic Authority & Three-States Detail

Three States: 13 Traffic Leaders (13%), 33 Page-2 Purgatory (33%), 54 Ghost Towns (54%). Traffic Leaders carry a median DR of 42; Ghost Towns a median DR of 7. The cohort is bottom-heavy: the median site draws 38 monthly organic visits, and a single large brokerage accounts for more than a third of all traffic in the study.

Appendix E · Two Premium Platforms, Side by Side

Two premium platforms, two independent deep dives one month apart, the same structural verdict. Agent Image is comparable on per-site technical debt and materially weaker on both AI visibility and organic authority. The June 2026 study proved the pattern was not unique to any single vendor; this one proves it persists, and deepens, on a second premium one.

Appendix F · The Vanity-Domain Trap

A structural pattern specific to this cohort deserves its own note, because it is both a measurement caveat and a finding.

Many sites in this study are built on a personal vanity domain, for example an agent's own name, while the agent's actual reputation, reviews, and press live under a brokerage or team domain. When a buyer asks an AI for the best agent in a market and the model does surface the person, it frequently cites the brokerage address, not the vanity site the platform built. Under our strict exact-domain ruler, that site scores zero even though the human was recommended.

We report the strict exact-domain rate as the headline because it mirrors our prior study and is the honest floor. But the deeper lesson is strategic: a single, authoritative, well-defined home for the entity is worth more to AI visibility than a beautiful vanity site that fragments it. Consolidation is a fix, and it is one the platform's default build works against.

Appendix G · Defensive Q&A

"These are just SEO warnings."

If they were minor, they would be evenly distributed. They are not. Half of everything concentrates in three categories that control AI interpretation, and that concentration coincides with 72% zero non-branded visibility across markets and brands.

"Our platform is technically sound."

Largely correct, and that is the point. The holistic technical build averages 877 of 1,000. The platform delivers a competent container. It does not deliver structured authority, entity definition, or AI-readable meaning, and the visibility numbers prove a sound container without engineered content behaves like an empty shell.

"AI visibility is not important yet."

We document the present, not the future. AI visibility is already measurable, already near zero for this cohort, and already excludes recognized market leaders. When a buyer asks an AI for local expertise today, these brands are not in the answer. And the trend runs one way: AI referral traffic has grown roughly sixteenfold in two years, with the leading engines up 200 to 320 percent year over year.

"You are just bashing a competitor's platform."

We named Agent Image because controlling for one platform is what makes the finding rigorous, and we credit its genuine strengths throughout, including a competently built container. We showed the identical pattern on a different premium platform one month earlier, and across many platforms in 2025. Every conclusion maps to a counted signal or an observed AI answer.