How 22 Dealer Website Platforms Handle AI Bot Access

Your website platform may let one AI assistant through and turn another away.

That is what stood out when CarGenius compared 22 dealer website platforms. On sampled Dealer.com sites, test requests identifying themselves as ChatGPT-User, Claude-User and Perplexity-User were refused, while requests using the Google-Agent name were not. Sampled DealerOn sites refused none of those four requests.

For dealers, the question is straightforward: if an AI assistant comes looking for your inventory, service hours or dealership details, how does your website platform respond?

The differences did not stop there. Some platforms behaved differently from one site to another, and published bot instructions did not always match the response to a request. The tables below separate actual test responses from written robots.txt rules and llms.txt availability.

Why bot access matters to a dealership

AI assistants can use information from several places, including search indexes, third-party websites and direct visits to a dealer's own pages. When an assistant retrieves your site, it may be looking for a vehicle's availability, service hours, location details or policies. If that request fails, the assistant loses one route to your current information. It may still find your dealership through other sources.

There is also more than one kind of bot. User-requested agents retrieve pages to answer a person's question. Separate crawlers, such as OAI-SearchBot, Claude-SearchBot, PerplexityBot and Googlebot, help their services discover content for search. Training bots serve another purpose, collecting material for possible use in AI model development. Dealerships should make their public inventory, service, and dealership information accessible to legitimate AI search crawlers and shopper-requested agents. Whether to allow AI companies to use that information for model training is a separate decision.

What we checked

Which AI bots did dealership websites refuse?

We tested two types of bot requests: agents that retrieve a page for a user, and crawlers that help AI services index pages for search. Both tables below show whether the sampled websites refused a request carrying the bot's name.

Dealer.com showed a consistent pattern: every sampled site refused ChatGPT-User, Claude-User and Perplexity-User requests, but not Google-Agent. Dealer Inspire showed a different pattern, refusing ChatGPT-User and Google-Agent requests while accepting the other two. The Dealer eProcess and Porsche OEM samples refused all four user-requested agents.

These refusals came from a test connection, not from the AI companies' verified networks. They are signs to investigate with a website provider, not proof that the official services cannot access those sites.

User-requested bots

These agents visit a page in response to a user's request, such as looking up a dealership or its current inventory. Google-Agent belongs in this group; it is not the same as Googlebot.

Search bots

These crawlers help search engines and AI search services find and index website pages. We checked OAI-SearchBot (OpenAI), Claude-SearchBot (Anthropic), PerplexityBot (Perplexity) and Googlebot (Google). Googlebot supports regular Google Search, including Google's AI Overviews and AI Mode. It is different from Google-Agent, which retrieves pages in response to a user's request.

Platform spotlight | Sokal + DataDome

Why Sokal refused every test request

All three Sokal-managed sites in our sample refused every scripted request, including one identifying itself as an ordinary browser. That is why Sokal appears as Refused in both direct-access tables. It does not tell us whether verified ChatGPT, Google, or other AI services can access those sites.

There is a reason for the strict screening. In a March 2025 customer case study, Sokal describes adopting DataDome to combat fraudulent ad traffic, website scraping, and denial-of-service attacks that strained its resources. Sokal reported a sharp reduction in unwanted traffic and less manual security work.

DataDome's documentation says unauthenticated automated requests are blocked by default, while recognized bots and AI agents can be authorized. Even legitimate shoppers may be asked to complete a bot-verification challenge before they can browse inventory or contact a dealership. That extra step adds friction. The Sokal-branded screen shown here illustrates the potential experience, but our study did not measure how often it happens.

Even human visitors may see a bot challenge. This Sokal-branded verification step can interrupt a shopping visit.

What dealers should ask: Can verified AI search and shopper-requested agents access our public inventory and service pages? How often do real customers encounter verification challenges? Strong security and useful AI access should be managed together.

What the robots.txt files tell different AI bots

A robots.txt file publishes instructions about which parts of a website automated visitors should access. That is different from a security system that actually allows, blocks or challenges a request. A website can publish an allowance for an agent and still refuse it at the security layer.

In our sample, Dealer.com, DealerOn and Team Velocity commonly published path-specific restrictions for user-requested and search bots. These were limits on selected parts of a site, not blanket blocks. Many other sites had no rule specifically naming an agent. That does not guarantee access, because a general rule may still apply.

Google-Agent-specific rules were uncommon across the platforms. Team Velocity published path-specific restrictions, while the Audi OEM sample returned an error when the survey tried to retrieve robots.txt. Google says its user-triggered fetchers generally ignore robots.txt. These published rules are therefore not a definitive test of Google-Agent access.

All means the same rule status appeared at every sampled site. Partial means statuses varied by site. Red means path restrictions worth reviewing, yellow means mixed results, green means an explicit allowance and gray means no clear agent-specific instruction.

User-requested bots

Search bots

These are the published instructions for the same four search crawlers tested above. A rule restricting selected URLs does not mean the whole website is blocked, and a missing bot-specific rule is not proof of access.

Training bots

Some AI companies use separate crawlers to collect material that may help train their models. These crawlers serve a different purpose from an assistant retrieving your inventory or service hours for a shopper. Your dealership can choose to allow AI-powered discovery while limiting the use of its website content for model training.

We checked the published robots.txt rules for GPTBot (OpenAI), ClaudeBot (Anthropic), Google-Extended (Google) and Meta-ExternalAgent (Meta). Google-Extended is a policy setting, not a separate crawler that visits the site. It helps control some Google AI training and grounding uses without determining whether a site appears in Google Search.

Colors indicate the site's stated access rule, not whether the platform is good or bad for AI visibility. A dealership may intentionally block training crawlers while allowing search and user-requested agents. Blocked means a full-site disallow rule was found; Restricted means selected paths were limited. The labels describe the published instructions for each named bot, not confirmed access by the real service.

Which platforms publish llms.txt?

An llms.txt file is a proposed way to provide AI tools with a concise guide to a website and links to useful information. It is not a replacement for accessible pages, and there is no established evidence that simply adding the file makes a dealership appear more often in AI recommendations.

Of the 22 platforms, 15 consistently published llms.txt across their sampled websites. One showed mixed results, and six had no file in the sample. Dealer eProcess illustrates why the file alone is not enough: the platform published llms.txt throughout its sample, yet those sites refused our simulated requests from all four user-requested agents. A text file cannot override the site's access controls.

All means every sampled site had an llms.txt file, Partial means some did, and None means none did. The table measures presence, not the quality of the information inside.

What happens when ChatGPT actually tries to access dealership inventory?

To understand how these technical limitations affect real-world AI shopping experiences, we ran a separate test using ChatGPT to explore inventory at Capital Ford Charlotte on October 23, 2026.

We asked ChatGPT to retrieve used vehicle inventory and navigate beyond the first page of results.

What we found was concerning.

In other words, ChatGPT could appear to be successfully accessing dealership inventory while actually working with older or incorrectly filtered information.

This illustrates an important distinction: An AI assistant finding your website doesn't necessarily mean it has access to your current inventory.

For shoppers asking about specific vehicles, pricing, or availability, that distinction matters. Inventory changes daily, and even a few days of delay can make information unreliable.

These observations don't establish that the dealership's website platform blocked ChatGPT from retrieving live pages. AI retrieval systems can rely on previously crawled content for several reasons, including retrieval limitations and website configuration.

But they highlight a broader challenge for automotive retail: making information discoverable is only part of the problem. AI agents also need reliable access to accurate, up-to-date inventory data.

As AI moves beyond recommending dealerships toward helping shoppers find and purchase specific vehicles, direct access to structured, current dealership data will become increasingly important.

Your website's settings matter more than the platform name

The survey found differences within some platforms. The Team Velocity sample included both sites that refused simulated agent requests and sites that accepted them. Jazel showed a similar split. That pattern is a reason to check your dealership's actual configuration, even if another store on the same platform appears accessible.

These checks also answer a narrower question than whether ChatGPT or another AI service will recommend your store. A website can be reachable and still have outdated inventory, thin service information or inconsistent dealership details. AI can also name a dealership using third-party information without retrieving its website directly. Technical access is one part of making your dealership's information available, not an AI visibility score.

For most dealers, the website provider controls hosting, security rules and bot settings. Your marketing team can identify the concern, but the provider is usually best placed to confirm whether legitimate AI traffic is actually getting through.

Four questions to ask your website provider

If your dealership appears in one of these platform groups, use the results to start a specific conversation with the team that manages your website.

About the data