Five takeaways from WordPress VIP and 97th Floor on finding, and fixing, the gap between what your client says and what AI tells buyers.
A 97th Floor client wanted to reach smaller venues. AI was describing the company in a way that would scare them off.
When Paxton Gray, CEO of 97th Floor, asked his team to check how AI described the enterprise venue-management company, the answer ran directly against its growth plan. AI described the company as established, clunky, and hard to use, and compared the brand to the IBM of venue-management software.
A buyer researching software for a smaller operation could encounter that description before ever reaching the company’s current website. If that were your client, where would you look first?
That example came from How AI Sees Your Brand, the first session in a three-part webinar series on AI discoverability, co-hosted by Gray and Jodi Cerretani, CMO of WordPress VIP. Gray has spent 20 years in SEO and calls AI search the biggest change he’s seen in that time. Half of B2B buyers now start their research in AI chat, according to G2 research, and two-thirds say it changed what they bought. Gray calls the distance between what a company wants AI to understand and what AI actually says the “reputation gap.” For agencies, the five recommendations from the session create a useful client framework: find what AI is seeing, fix what is wrong, and check it again later.
Here are the five places to start.
1. Old Properties Can Keep Telling the Old Story
A redesign doesn’t erase the old internet. Microsites, legacy subdomains, product PDFs, and campaign properties can keep describing a company the way it stopped describing itself years ago.
During the webinar, Cerretani described this as the total information environment: the live site, everything still parked on old subdomains, microsites, press coverage, Reddit threads, and other public material associated with the company.
WordPress VIP measured AI comprehension across six sites and found scores ranging from 59% to 88%. The properties were telling different versions of the same company story. If six properties tell six versions of the story, AI has to decide which one to believe. AI averages what it finds, and an average brand is a diluted brand.
Start with an inventory. Which domains, subdomains, microsites, PDFs, and campaign pages are still public? What do they say about the company’s products, audience, and positioning?
Most clients have more old material in circulation than anyone on the current team remembers publishing, and finding it is the fastest way to turn a vague worry into a list of fixes.
2. Check Whether AI Can Reach the Pages That Matter
Publishing a page does not mean an AI crawler can retrieve it.
Across 28 million server requests for one pharmaceutical company, WordPress VIP found a group of product pages with 230,000 human visits and only six AI crawler visits.
The causes were ordinary. The pages were missing from the sitemap, and some critical content lived inside JavaScript that appeared as empty boxes to AI crawlers.
Check whether priority URLs appear in the sitemap, whether AI crawlers request them, and whether the important copy remains available when JavaScript doesn’t run. Add a text fallback behind every JavaScript component. If the crawler never sees the new product page, that page can’t correct the old story.
3. A Page Can Look Right and Still Be Hard for AI to Understand
Readable to a person is not the same as legible to a crawler.
Cerretani describes an AI crawler as a brilliantly fast reader that’s also completely literal. It doesn’t read size, placement, or color the way a person does. It reads the structure underneath like headings, semantic HTML, schema, navigation, hierarchy.
During one enterprise assessment, WordPress VIP found 16 product pages with no schema markup. While the pages looked fine to human visitors, crawlers had fewer clues about what the product was and how the page was organized.
That review happens at the template level. Check the H1, the heading hierarchy, the semantic HTML, the schema, and whether related page types are built consistently. One template fix can lift an entire class of product pages at once, and the same work counts as an accessibility win.
4. Your Analytics Are Not Showing You Everything
Most client dashboards were designed to measure human traffic.
Google Analytics, Adobe Analytics, and similar tools generally rely on JavaScript running in a browser. AI crawlers often make server requests without executing that JavaScript. The dashboard may never see them but the server logs do.
During one enterprise audit, WordPress VIP found a regulatory-compliance plugin rejecting roughly 15% of ChatGPT requests, about 68,000 in the measured period. The plugin treated AI crawlers like human visitors from blocked countries. WordPress VIP then ran the same audit on its own site and found a similar problem. The server layer was rejecting one of Anthropic’s crawlers, on a site that scores well for technical AI readiness.
Neither started as an AI problem. Both came from ordinary infrastructure and security decisions that nobody thought to revisit.
Ask to see the server logs. Look at which crawlers arrive, which URLs they request, which requests succeed, and where access stops.
This is also where agency coordination matters. Marketing may own the question, while developers, WebOps, security, infrastructure teams, or the hosting provider own the evidence and the fix. Agencies often already work across both those teams.
5. Don’t Make This a One-Time Audit
AI discoverability is not something you fix once. Teams keep publishing new pages while old assets stay up, plugins get installed, security rules change, templates evolve, and the models themselves keep moving. Cerretani also points out that AI systems have a strong recency bias, so content goes stale faster than it used to.
WordPress VIP repeats this type of assessment about every 90 days, and some customers review their environment more often. The session’s closing recommendation was to name an owner, retest on a schedule, and keep the teams involved talking to each other.
For agencies, that makes the work repeatable—and turns a project into a program. A recurring review can answer four questions: What can AI reach? Does the site clearly communicate the company’s current positioning? Where are technical rules interrupting access? What changed since the last review?
Agencies that already understand the client’s publishing workflow, templates, hosting environment, and approval process are well positioned to own that review.
What to Sell, and How to Prove It Worked
The audit is where you start, and it sells itself once you show a client the microsite they forgot about. From there, the fixes become a scoped project covering the sitemap, JavaScript fallbacks, schema, templates, and whatever the server logs turn up. When that project closes, the quarterly review keeps you on retainer.
Clients want to know how they’ll prove any of it worked, and right now nobody can draw a straight line from an AI mention to revenue. Gray splits the measures in two, which is the most useful thing you can take into that conversation.
Leading indicators move first: brand mentions, citations, and share of voice against competitors. If those are climbing, it’s working, and you can report that inside 90 days.
Lagging indicators follow: LLM referral traffic, branded search, direct traffic, conversion rate, and time to conversion. Gray’s point is that people who research in AI chat arrive ready to buy, so conversion should also get faster.
He compares it to a billboard. You can’t trace a billboard to a sale either. You watch whether more people search your name, come direct, and convert better. That framing lets a CMO act without waiting for attribution that doesn’t exist yet.
Gray’s read is that the brands that win will be the ones willing to move before the dots connect, and that plenty of teams will wait instead. That’s the argument to make, and the reason to make it now.
You’ll know the demand is real when you hear how the request arrives. It used to be that the CEO googled the company name and didn’t see it. Now the CEO asks a chatbot and doesn’t come up. Cerretani says the question is reaching boards, which means your marketing contact may be getting asked about this from above without any way to answer.
Five Questions to Bring to Your Next Client Review
- Which domains, subdomains, microsites, legacy pages, and PDFs still describe this company?
- Can AI crawlers reach the product and service pages the client wants buyers to find?
- What do the headings, semantic HTML, and schema tell a crawler about those pages?
- Which AI crawler requests appear in the server logs, and what do those logs reveal about access and blocks?
- Who owns the next review, and when will it happen?
Why WordPress Agencies Start Ahead
As Gray put it during the session, AI was trained on WordPress. It makes up a huge share of the web the models learned from, which makes it a strong technical base to build on. Pair that with content written for the questions buyers actually ask, and you have most of what AI search rewards.
For Gray’s venue-management client, the growth goal was clear and AI was working against it. The useful question is why AI says what it says, and what you can change.
Pick one client whose AI answers look wrong. Check the old properties, the priority pages, the markup, and the server logs. You’ll find something worth fixing.
Watch How AI Sees Your Brand to hear Jodi Cerretani and Paxton Gray unpack the AI reputation gap, walk through real enterprise audit findings, explain how AI discoverability changes measurement, and share five actions teams can take now. It’s the first session in our three-part AI discoverability webinar series.
Automattic 