Where consumers go to trust a brand: how AI decides whether to cite you

When someone wants to know whether a brand is trustworthy, searching the brand name up isn’t the first option anymore. They ask an LLM to compare it against alternatives, check Reddit, look at some review aggregators, or even see whether it comes up through a dedicated AI search engine.

As such, the brand’s own website might be the fifth or sixth touchpoint, if it appears at all. This means you need to learn how AI systems cite you while still maintaining the consumer trust you’ve built over time.

The 5 channels brand trust forms before a visitor reaches your website

Brand trust used to run through a single channel: a search result, a click, a landing page, maybe a contact form or download. It now passes through five. A big difference is that your website is usually the last of them rather than the first:

• AI-generated answers. ChatGPT has passed 900 million weekly active users. Google AI Overviews now trigger on nearly half of all tracked search queries.
• Community platforms. Reddit threads, niche forums, and LinkedIn discussions feed directly into what those AI answers say.
• Review and comparison platforms. G2, Trustpilot, and Capterra build the kind of cross-referenced agreement a model can act on with confidence.
• Third-party editorial coverage. Independent publications and industry sites carry more weight with some platforms than a brand’s own pages ever will.
• The brand’s own website. This is the last stop, not the first. Now, this is the point where a visitor comes to you once they’ve already formed an opinion.

To go deeper, AI Overview coverage figures are up 58%, which means the channel a brand can least directly control is also the one growing fastest. This also applies to conversions too: AI-referred visitors convert to sign-ups at roughly 11 times the rate of traditional search traffic. It makes a citation function closer to a warm lead than a passing mention.

Combined, these channels now feed into how an AI platform decides who to cite, so a weakness in one undermines the rest. As such, a brand with excellent content but no community presence or review coverage is judged more of this gap than on content.

Three AI platforms decide who to cite in different ways

Most coverage of AI visibility treats “being cited” as a single goal, as if optimizing for ChatGPT automatically helps a page show up in Perplexity or Google’s AI Overviews too. However, each platform retrieves from a different index or favors a different mix of sources.

So, a brand can dominate one platform’s citations but disappear from another for the exact same query. Because the two systems are asking different questions, it helps to understand what each platform looks for.

ChatGPT favors consensus across independent sources

ChatGPT’s citations concentrate around a small number of sources. For instance, Wikipedia alone accounts for nearly half of citations among its top 10 most-cited sources. This points to a strong preference for encyclopedic, cross-referenced information over a brand’s own claims about itself.

Reddit sits a distant second, as it only accounts for around 10% of that same top-10 share. It means a well-regarded thread carries more weight with ChatGPT than most brands assume.

In short, ChatGPT is looking for the same facts to show up more than once from sources that don’t answer to the brand. For instance:

• A G2 or Capterra profile stating the same core positioning as the brand’s own site.
• A Reddit thread or forum discussion where the brand gets recommended unprompted, not in a marketing brief.
• Independent press or industry coverage that repeats the same facts without being asked to.

If you’re chasing ChatGPT citations by polishing your own homepage, you’re solving the wrong problem. Instead, the model is short on outside confirmation and verification about your own claims.

Perplexity favors community discussion and fresh content

Perplexity runs on a different logic entirely as it builds around live retrieval rather than a fixed index. It shows up in two ways:

• Heavy community concentration. Almost 50% of its top citations come from Reddit, which is flipped in comparison to ChatGPT and far ahead of any other source type.
• A steep freshness curve. Content updated within the last 30 days gets cited at roughly 82%. Content over a year old is cited at only 37%.

Based only on this, both Perplexity and ChatGPT reward different kinds of focus. Because Perplexity runs a live web query for nearly every prompt, a brand that revisits its most important pages can see changes reflected within weeks. However, an important channel for many in AI Overviews can take months to catch up.

Google AI Overviews still runs on ranking signals but is less predictable than it looks

AI Overviews builds on Google’s existing ranking systems, including E-E-A-T and Core Web Vitals, to add an AI summarizing layer on top.

This link to ranking has typically been a reliable predictor of citation, but one stat in particular highlights how it no longer is: only 38% of pages cited in Google AI Overviews also rank in the top 10 organic results for the same query. This is down sharply from around 76% a few months earlier. It means while ranking well still helps, it doesn’t guarantee a citation.

To tie these together further, there’s a wide gap between all of the platforms and what they ‘optimize’ for:

• Only around 11% of domains get cited by both ChatGPT and Perplexity for the same query.
• Roughly 71% of cited sources show up only on either ChatGPT or Perplexity.

In a nutshell, treating AI visibility as a single target isn’t a realistic goal. However, being mindful of your own site’s performance can boost the effort you put into each platform.

A slow or blocked website loses citations before your content is ever judged

Even in cases where you’ve done the upstream work well and developed content that’s a strong candidate for citation, there’s more work to do. For instance, consider an AI crawler timing out when visiting your site. Despite the work you’ve already undertaken, the citation won’t happen anyway.

It’s a common point of failure because AI crawlers behave differently to how most brands assume a search engine crawler would:

• They fetch pages live rather than pulling from a cached index, so a slow server response shows up as a failed (rather than delayed) visit.
• They time out faster than Googlebot, which means a page that eventually loads for a human may never load for a crawler.
• They treat an error as a reason to deprioritize the source, so a single bad request can affect whether the crawler comes back to try again.

Kinsta customer Mekari ran into this problem before it fully understood the connection. Its team traced weak search visibility back to low crawl requests and high server response times within Google Search Console.

As Mekari’s analytics team explained:

We compete on transactional keywords. If our website is not crawled easily by Google bots, whether it is because of slow response time or page load time, then we will not be visible to the target audience.

After moving to Kinsta, Mekari’s server response time fell by close to half, crawl requests nearly tripled, and search traffic grew by a third. This happened because we reacted to crawler behavior rather than content changes.

Finally, security settings can also create the same blind spots. For instance, switching on settings to block AI crawlers for bandwidth or security reasons removes you from AI citation consideration. It’s a trade-off that is sometimes right, but should always be a deliberate choice.

How to check your citation readiness inside MyKinsta

MyKinsta brings uptime, page performance, and crawler access into one dashboard. Here’s how to use each area to diagnose a specific citation problem.

Start with uptime and error monitoring

A citation depends on the moment a crawler happens to visit over the average state of your site, so the interval between checks matters more than for a human visitor. MyKinsta checks every site up to 480 times a day, across all plans.

Enable monitoring alerts under User Settings > Notifications. Once monitoring flags an issue after three consecutive failed checks, you’ll get an email.

If an alert fires, go to Analytics > Response first. The response code breakdown shows whether you’re dealing with a 5xx server error or a 4xx resource issue. That distinction tells you where to investigate before you start changing anything.

Trace slow pages back to their actual cause

A slow page costs a citation the same way an outage does, just less visibly. For example, if nothing returns an error or gets flagged, the page will lose out to a faster source answering the same question.

Kinsta’s APM tool lets you monitor across a window of two to 24 hours. While it’s running, you can reproduce the slow request and then read the results across four views:

• Transactions show which specific page or endpoint is actually slow, rather than a site-wide average that hides the problem.
• WordPress breaks the delay down by plugin and theme, so one misbehaving plugin doesn’t get lost among the rest.
• Database shows slow or repeated queries, such as a plugin conflict or a bloated options table.
• External flags third-party API and service calls, often the bottleneck on a page that otherwise looks fine.

Finding the cause is the most important here, since a caching fix and a code fix solve different problems. Treating one as the other wastes the time you were trying to save.

Confirm AI crawlers can reach the site

Kinsta’s Bot Protection always lets verified search engine crawlers such as Googlebot and Bingbot through. However, AI crawlers don’t get the same automatic pass. As such, you need to make a check here rather than an assumption:

• The dedicated Block AI crawlers toggle lets you opt an AI crawler out entirely in one switch.
• An Excessive rate AI crawlers category reclassifies any verified bot generating unusually high request volumes, challenging it at stricter protection settings.
• Bot traffic analytics break all site traffic down by category: verified bots, AI crawlers, automated traffic, and likely humans.

If AI crawler traffic is being blocked and that wasn’t intentional, the Request breakdown will show it. Change the protection level or toggle the Block AI crawlers setting off, then check again after 24 hours to confirm traffic is flowing through. For sites where crawler load is genuinely straining server resources, the right answer may be to keep the block, but check the Bot traffic analytics first so you know what you’re actually blocking before deciding.

These three checks together tell you whether your site is structurally capable of being cited. Any problem found here is an infrastructure problem, not a content problem, and the data from MyKinsta is something you can show directly to clients rather than ask them to take your word for it.

What agencies should tell clients who ask why traffic is flat

Agencies already have a version of this whole conversation, usually framed around traffic concerns despite a consistent publishing schedule. From your end, the honest answer is that brand trust begins away from a website.

Consider the number of times a customer has already ‘met’ your brand through the various other channels, then consider what that customer is looking for. If your (or your client’s) site is slow or unreliable, it undermines the trust that has hopefully been earned elsewhere.

It also means there are three questions worth bringing to a client meeting instead of a stats report:

• Where is the brand being cited today and on which platform?
• Where is the brand missing and what is it costing you?
• What does the infrastructure need to close any gaps?

Adapting Social, a digital marketing agency managing more than 50 client sites, brings this type of framing to its own client conversations. After standardizing on faster, more reliable hosting, the team sees rises in search traffic for many of its clients. It’s proof that the technical half of this conversation moves the number just as much as content does.

Brand trust is earned in the ecosystem and lost in the infrastructure

Brand trust now forms upstream of your website. Your site’s job is no longer just where trust is established, but where trust is confirmed.

A site that’s slow, intermittently unavailable, or inadvertently blocking AI crawlers fails that confirmation test. Before revisiting a content strategy, check whether uptime monitoring is on, whether a slow page is costing citations, and whether Bot Protection settings are doing what you actually intend. All three checks take under five minutes in MyKinsta.

Kinsta’s managed hosting for WordPress keeps uptime monitoring, performance infrastructure, and crawler access working as one layer rather than three separate things to manage.

The post Where consumers go to trust a brand: how AI decides whether to cite you appeared first on Kinsta®.