Four takeaways from WordPress VIP and 97th Floor on turning an AI audit into a plan your client can act on.
One of 97th Floor’s clients, a cruise line, wanted to be the brand AI recommends to anyone planning a trip to Alaska. For a long time, it traded the top spot with competitors.
97th Floor removed the pages that didn’t fit and built new ones around the questions Alaska travelers ask, like when to go, which animals they’ll see, and which excursions are worth booking. Each piece matched a stage of the buyer’s journey. Once the strategy was in place, the client pulled ahead and became the most cited cruise line in the world for Alaskan cruises.
That story came from Build an AI-Ready Discoverability Strategy, the second session in a three-part webinar series on AI discoverability. Jodi Cerretani, CMO of WordPress VIP, and Paxton Gray, CEO of 97th Floor, co-hosted it. Session one covered how to find out what AI says about a brand. Session two picks up after the audit and covers what to fix first and who should own the work.
Gray and Cerretani group the work into four areas: brand, architecture, content, and third-party validation. This session focused on architecture, content, and third-party validation, along with how to prioritize and staff the work. The takeaways below are the ones most useful to agencies planning client work.
1. Build Consistency Into the System
In Cerretani’s library comparison, information architecture is the floor plan that shows visitors where each section is. Structured content is the label on each book, and governance is the librarian who keeps books shelved and labeled correctly as the collection grows.
Most content systems were built before ChatGPT, and WordPress VIP sees them break in the same places. It recommends three fixes:
- Name things the same way everywhere. If one page says “WordPress VIP” and another says “enterprise WordPress,” AI may read them as two different things. Models are getting better at connecting the two, but it’s still a risk.
- Model content once and reuse it. Pricing tiers, product names, feature names, and executive names belong in structured fields that update everywhere from one source. Free-text blocks that someone retypes on every page drift out of sync.
- Put machine-readable tags in the template. If schema depends on someone remembering to add it, some pages will miss it.
Cerretani suggests asking every client whether their website software handles this automatically or depends on people remembering. If it depends on people, expect gaps.
To keep those fixes in place over time, write down your standards for headings, titles, metadata, schema, and naming. Add automated checks to the publishing tools, similar to spell check, and name one owner for each domain.
Most of this happens at the template and content-model level, where many agencies already work for their clients. A fix there updates every page built on that template.
2. Plan Around Topics Instead of Keywords
For 20 years, SEO meant getting a single page to rank in Google’s 10 blue links. Gray says AI search works differently, and one of the most common mistakes he sees is treating prompts like keywords.
When someone types a long, detailed prompt, AI breaks it into several searches, a process Gray calls the query fan-out. A site that answers those follow-up questions across several related pages has a better chance of showing up. The more authority it builds on a topic, the more likely AI is to recommend the brand.
Gray’s team starts with audience research, reading Reddit and other communities to learn the words and problems buyers bring up. From there, they build personas around pain points, motivations, language, and questions. The content plan comes from those personas: a main page on the core topic, with narrower supporting pages that each address one problem and link back.
For agencies, that means a content retainer with more pieces, each narrower and tied to what buyers ask.
3. Cut Content That Teaches AI the Wrong Thing
Gray says many enterprise brands carry 20–30 years of content behind them, which he compares to a wake they’ve been surfing. Old pages still describe discontinued products or past positioning, and AI still reads them.
97th Floor runs every new client’s content through a matrix with three questions:
- Does the page get traffic?
- How does it perform on engagement, time on page, click-through rate, and bounce rate?
- Does it match the business today?
Teams tend to skip the third question, but a page with good traffic and engagement may still need to come down if it describes the company in a way that no longer fits. Removing those pages gives AI a clearer picture of who the company is and who it serves. That was the first step for the cruise line.
When an attendee asked how to sort through thousands of old blog posts, Gray said he starts with the AI sentiment analysis and checks the citations. If AI says something wrong about the brand and cites the brand’s own site, those pages are quick wins. After that, he works through the rest sorted by traffic and engagement, highest first. He admits this part is slow, but sorting this way means the highest-impact pages get reviewed first.
Cerretani said her team sees something similar with AI tools internally, where giving a model more context can make its answers worse. She sees content the same way, with clarity mattering more than volume.
4. Find Out Where AI Learns About Your Client
Gray recommends a brand sentiment analysis to find out what AI thinks about a client’s brand and which sources it draws from. The client’s own site is one of those sources, but the analysis also shows the language AI uses, where it places the company in the market, and which outside sites shape those answers.
Reddit, YouTube, and LinkedIn get most of the attention, and they do influence AI answers. Gray says the most cited sources vary by industry. For one 97th Floor client, JazzHR was the third most cited site, so the plan included building a stronger presence there and on Lever.
When someone asked whether Quora still matters, Gray said it depends on whether AI pulls from it for your brand. If it does, go correct or add to what’s there.
Larger brands can also use partner marketplaces. For now, AI treats listings across multiple marketplaces as third-party validation of what a brand does and who it serves, which gives partner marketing teams a direct way to influence AI answers. Cerretani added that the naming consistency from the technical section applies to these listings too.
Sequence the Work and Name an Owner
Cerretani recommends starting with quick wins that take a few weeks:
- Fix wrong tags on the most important pages.
- Unblock AI crawlers that got locked out by accident.
- Patch holes in the sitemap.
- Clean up places where the story conflicts across domains or pages.
The longer-term work includes rebuilding the content model and tagging system and deciding whether the current platform can automate what the client needs. It also covers building authority on core topics through on-site content and third parties, and writing governance rules. Once that’s in place, the team should test, fix, retest, and report on a regular schedule.
Marketing owns the narrative, the content team tracks buyer questions, and SEO maps the fan-out against competitors. Web teams manage templates and site hierarchy, engineering handles performance, security sets crawler policy, and a governance team enforces standards. In Cerretani’s experience, many companies have no single person accountable across all of those teams. She says the owner can come from any of them, as long as one person is responsible.
In the live poll, most attendees said they were working on technical and content foundations, and only 5% said they had a clear owner.
Where Agencies Fit
Agencies already work across a client’s templates, content, and hosting. They’re well placed to fill that ownership gap or run the process on the client’s behalf.
The work starts with the audit from session one, followed by a quick-win sprint. A larger project comes next, covering the content model, templates, content pruning, topic clusters, and third-party presence. After that, a recurring review keeps the work current.
Cerretani says boards now ask CMOs what they’re doing about AI discoverability, since organic traffic and unpaid pipeline have been declining for many brands. She doesn’t recall boards asking the same about SEO, which means your marketing contact is probably already fielding this question.
Six Questions for Your Next Client Planning Session
- Does the client name its products, people, and industry the same way on every page and property?
- Do key facts like pricing tiers, product names, and leadership come from structured fields or free text?
- Is schema built into the templates, or added by hand?
- Which old pages still get traffic but no longer match the business?
- Which third-party sites does AI cite when it describes this client?
- Who owns AI discoverability, and when is the next review?
Why WordPress Agencies Start Ahead
Gray calls WordPress a strong technical foundation for this work, since so much of the web that trained AI runs on it. Templates, blocks, and custom fields let agencies build consistent naming, structured content, and schema into the site so it doesn’t depend on someone remembering.
Cerretani ended the session by stressing that infrastructure and content depend on each other. Good infrastructure makes a brand readable to AI, and good content gives AI a reason to cite and recommend it. If AI can’t read the site, even the best content won’t get cited.
A good place to start is one client whose AI answers look off. Work through the quick-win list with them and see what you find.
Watch Build an AI-Ready Discoverability Strategy to hear Jodi Cerretani and Paxton Gray walk through the technical foundation, content strategy, third-party validation, and ownership model behind AI visibility. Session three, on how to measure it, is on October 20. Register here.
