Is SEO dead because of AI Overviews? No. But the way you report on it needs a serious update, and honestly, so does the conversation you’re having with your leadership team about it.
TL;DR: If your traffic dropped and you’re not sure whether to blame AI Overviews, your SEO strategy, or a shrinking market, don’t guess. Build a site specific click through rate curve, check your rankings and share of voice before you touch the traffic number, confirm whether search demand actually declined, track AI Overview impressions as a brand signal, and run a quick audit of how LLMs describe your brand. Bring all of it into one meeting instead of one anxious “traffic is down” email.
This post is based on an Ahrefs live session featuring Lidia Infante. Full credit for the frameworks, methodology, and data walkthrough goes to Ahrefs and Lidia Infante. You can watch the full session here: Traffic is down, AI Overviews are up: What do you tell the C-Suite?
FAIR WARNING: this video is super technical…you might just want to read the summary below. :roll:
I missed the live version of this webinar. It aired at 4am my time here in Texas, and even this coffee loving early bird has limits! So I watched the replay instead. Lidia Infante, an SEO leader and marketing strategist with 10 years of experience in organic growth, currently at Ahrefs and previously at SurveyMonkey. It was worth staying up late for the recording, so I’m breaking it down here.
Webinar summary:
The Story Executives Are Hearing Isn’t the Full Picture
Revenue is down for a lot of companies lately, growth has slowed, and it’s easy for “SEO is dead” to become the explanation of choice, especially when that story is circulating on LinkedIn. Infante pushed back on this framing hard. Search volume hasn’t collapsed across the board. What’s changed is where clicks go once someone searches.
AI Overviews took a slow, gradual erosion in organic click through rate and turned it into a sharp one. As AI Overviews and AI Mode keep expanding into more queries, the pressure on organic traffic keeps building. That’s a real structural shift, but it’s a different problem than “nobody is searching anymore,” and it calls for a different response.
Step 1: Build an AIO vs. Non-AIO Click Through Rate (CTR) Curve
Rather than accepting a generic industry statistic about AI Overview impact, Infante walked through building a site specific click through rate curve using Ahrefs Site Explorer and Google Search Console data. The process, in broad strokes:
Pull your organic keywords from Ahrefs, filtered to positions 1 through 10 (anything beyond that is mostly noise for this purpose)
Normalize keyword formatting so your Ahrefs and GSC exports actually match up
Flag each keyword as AIO or non-AIO based on Ahrefs’ search features data, and flag branded vs. non-branded separately, since brand recognition affects click behavior independently of AI Overviews
Bring in your GSC clicks and impressions, then calculate CTR by position for each bucket
Apply a leverage check so no single keyword is skewing your curve, generally keeping any one keyword’s influence under roughly 30 percent
The result is a curve that shows, in your own data, how much value a top ranking position has lost once an AI Overview appears. In the example used during the session, position one lost about 13 percent of its expected clicks once an AI Overview showed up for that query. Infante noted she’s seen swings as steep as 82 percent on other sites, so the impact varies widely by industry and query type.
Step 2: Check Rankings and Share of Voice Before Traffic
Before traffic became unreliable as a single KPI, position was the leading predictor of it. Infante’s recommendation is straightforward: check whether your rankings and keyword coverage have actually declined before concluding that your SEO program has stopped working. If positions are stable or improving, that’s evidence the strategy is sound and the rules around what traffic converts to have simply changed.
She also pointed to share of voice as an underused metric worth bringing into the boardroom. Unlike traffic, share of voice isn’t affected by whether an AI Overview shows up or not, since it measures your visibility relative to competitors across a set of tracked keywords. It’s a clean way to set targets and rally a team around taking market share back from competitors, and it holds up well when a CFO asks for a single number that represents SEO performance.
Sometimes search volume for a given topic genuinely is declining, and it’s worth checking rather than assuming. Infante suggested pulling tracked keywords into Ahrefs Keywords Explorer to calculate a search volume growth rate by topic or product line over time. In her own case study using Typeform style keyword sets, tool related searches dropped in volume year over year, while use case and “how to” searches around the same topic actually increased. That kind of nuance changes where you invest content effort, and it’s a much stronger answer to “people aren’t searching anymore” than a blanket yes or no.
Step 4: Use AI Overview and AI Mode Impressions as a Brand Signal
Even when an AI Overview doesn’t send a click, it can still be doing work for your brand. Infante recommended tracking AI Overview impressions (available through tools like Ahrefs Brand Radar) and AI Mode impressions in Google Search Console as indirect evidence that demand is intact. Her observation: a jump in AI Overview impressions is often followed by an increase in branded search a few weeks later, which suggests people are learning about a brand through the AI Overview and then coming back to search for it by name.
Step 5: Run a Quick LLM Visibility Audit
To walk into a stakeholder meeting ready for questions about how the brand shows up in AI tools, Infante suggested a simple exercise: ask each LLM platform you care about the same six questions, repeated five times each, and look for consistent patterns. Questions should cover who your brand is, what makes it unique, what your best product is, and how it compares to competitors. Running the same questions against competitors adds a useful comparative angle for leadership conversations.
She also recommended log file analysis (Ahrefs offers a free bot analytics tool in beta) to see how often LLM crawlers are actually retrieving your content, separate from whether they’re sending traffic.
Meeting with the Boss
What to bring to the meeting: one slide covering the CTR curve, ranking trends, and share of voice, a second slide covering search volume trends and AI impression data as a rebuttal to the “nobody searches anymore” narrative, and notes from the quick LLM audit.
The goal isn’t to dismiss the impact AI Overviews are having. It’s to replace a vague, anxious conversation with a specific, data backed one that shows leadership exactly what changed and what to do next.
Site: FamilyCaregiversOnline.net, a free, ad-free, publicly funded resource for family caregivers, built on Divi
Starting AI Visibility Score: 65/100
Ending AI Visibility Score: 91/100
What changed: no llms.txt or llms-full.txt, no AI usage signals in robots.txt, a duplicated headline, missing Twitter/X cards, no markdown version for AI agents, and a caching hiccup quietly holding the score down at the finish line
Time to fix: one focused audit, worked through page by page and setting by setting
The website Family Caregivers Online exists for one reason: to make sure the people quietly caring for a parent, a spouse, a sibling (while also working, raising their own kids, running their own lives) have somewhere free and trustworthy to turn. No ads. No upsells. Just a publicly funded website doing exactly what caregivers do for everyone else: showing up, consistently, asking nothing in return.
This website is doing good, important work for its community. It just hadn’t had anyone check its own vitals in some time.
The AI Visibility Audit for a Nonprofit
I ran this AI Visibility audit for my client’s nonprofit website to see if tools like ChatGPT, Claude, and Perplexity can actually find the site, read it, understand what it offers, and confidently point someone toward it. Starting score: 65 out of 100. Not neglected. Just under-cared-for in the specific ways that don’t show up unless you go looking.
Here’s what the assessment turned up:
No care plan on file. No llms.txt or llms-full.txt, the files that tell AI tools what a site is, who it’s for, and where to find its resources. Without them, AI tools were left guessing at a site whose whole purpose is to not make people guess.
No documented consent for AI visits. The robots.txt file had no Content Signals: the emerging standard that tells AI crawlers explicitly whether they’re welcome to use a site’s content for search and citation versus training. Silence isn’t the same as an answer.
A confused intake form. Two H1 headings competing on the homepage, a leftover of a Divi responsive design pattern, where a desktop hero and a mobile hero had both been marked as the page’s primary headline. Harmless to a human visitor. Confusing to a crawler trying to figure out what the page is actually about.
No way to properly introduce itself. Twitter/X Card tags were missing, so when the site’s resources got shared on social media, there was no proper preview card, just a bare link doing none of the work of explaining what someone was about to click into.
No easy-read version for visitors who need one. No markdown delivery for AI agents, meaning AI tools had to wade through full page HTML to read an article instead of getting a clean, direct version of the content.
None of this was visible to a family caregiver landing on the homepage at midnight looking for help. All of it mattered to whether an AI tool would ever reliably find and recommend this site to the person who needed it.
The Care Plan
Here’s what actually got treated, one item at a time:
Wrote the care plan. Generated llms.txt (a structural index of the site’s core pages) and llms-full.txt (an expanded reference with mission, funders, and program details), so AI tools reading the site now get an intentional summary instead of having to piece one together themselves.
Documented consent for AI visits. Added Content Signals to robots.txt (search=yes, ai-input=yes, ai-train=no), explicitly welcoming AI tools to read and cite the site’s content for real-time answers, while opting out of it being used to train foundation models.
Resolved the intake mix-up. Traced the duplicate H1 to Divi’s separate desktop and mobile hero blocks and brought the mobile heading down to an H2 in the Divi Builder, so the page has one clear, unambiguous headline again.
Enabled the introduction card. Turned on Twitter/X Card output in Yoast’s social sharing settings and set explicit Twitter title and description fields on the homepage, since Yoast quietly skips writing those tags when they’d just duplicate the Open Graph ones.
Added the easy-read version. Enabled VigIA’s built-in Markdown-for-Agents feature across posts and pages, so any AI tool that wants a clean, direct version of an article (instead of full page HTML) can request one automatically.
Caught a filing error along the way. Both new AI files initially displayed with garbled special characters (em dashes and accented letters turning into gibberish) because the server wasn’t declaring a character set. Fixed it by adding a UTF-8 byte-order-mark directly to the files, so they’d render correctly no matter what the server did or didn’t announce.
Checked vitals on the way out. Homepage response time briefly spiked to over a full second right after all the changes went in, a caching side effect of the edits, not a real problem. A cache purge brought it back down to 33 milliseconds.
The AI Visibility Audit for a Nonprofit Results
Score progression over the audit:
Initial audit ……………………………………………… 65/100
After llms.txt, llms-full.txt, and Content Signals …… 82/100
After Twitter Cards, the duplicate H1 fix, and Markdown for Agents … 86/100 (masked by a temporary caching hiccup)
Final, after clearing the cache …………………………… 91/100 (A+)
Before and after, by category:
Category
Before
After
Access and AI discovery
20/37
37/37
Structured data and semantic context
18/25
21/25
Content structure and readability
13/20
15/20
AI interaction and distribution
4/8
8/8
Access performance
10/10
10/10
Same site, same content. No redesign. Just the kind of check-in this site has been giving to family caregivers all along, finally turned around and pointed at itself.
What I Learned
There’s nothing exotic about what was wrong here. It’s the ordinary, invisible residue of a site whose entire team has always been focused outward, on the caregivers who need it, rather than on the site’s own technical upkeep. That’s not a failure. It’s just what happens when an organization’s whole reason for existing is taking care of other people first.
It’s also exactly the blind spot an AI visibility audit is built to catch. If your organization’s website has been quietly doing good work for a while without anyone checking whether AI tools can actually find and recommend it, that’s the same nagging feeling I hear from a lot of clients, nonprofits especially, where there’s rarely time or budget for this specific kind of maintenance.
Book a free strategy session and let’s run an AI Visibility Audit on your site. Even the sites that take care of everyone else deserve a little care of their own.
A plain-English guide to Answer Engine Optimization, why it matters now, and how small businesses can get found by AI tools like ChatGPT.
TL;DR AEO stands for Answer Engine Optimization, the practice of making sure AI tools like Google’s AI Overviews, ChatGPT, Copilot, and the other AI tools can find, understand, and recommend your business. It doesn’t replace SEO, it builds on it. AI-driven traffic tends to convert at a much higher rate than traditional search traffic, because the people asking AI a question are often further along in deciding who to hire. If you’ve been putting off dealing with any of this, the good news is you probably don’t need to start from zero. Most small businesses already have the foundation, they just haven’t optimized it yet.
If you’ve heard the term AEO floating around and felt a little dread, you’re not alone. I talk to business owners every week who know something has shifted in how people search, but they’d rather not think about it too hard, because it sounds like one more thing to learn, one more thing to fear falling behind on. I get it. But burying your head in the sand doesn’t stop your traffic from changing. It just means you find out the hard way, usually when you notice fewer calls coming in and can’t figure out why.
So let’s talk about what AEO actually is, in plain English, and why it’s genuinely good news for a business like yours.
What AEO Actually Means
Answer Engine Optimization is the practice of making your website visible and useful to the AI tools that are now answering people’s questions directly, instead of just handing them a list of links. That includes Google’s AI Overviews, ChatGPT, Claude, Gemini, and Copilot. You might also hear it called GEO (Generative Engine Optimization) or LLMO (Large Language Model Optimization). Different label, same idea: instead of competing for a spot on a results page, you’re trying to be the business the AI actually mentions by name.
I’ve been digging into a new AEO training from Ahrefs, taught by Sam Oh, and a few numbers from the first lesson explain why this matters right now, not in some vague future.
As of December 2025, when Google’s AI Overviews began to show up on a search…
It cut the click-through rate for the number one ranking result by 58 percent. That means a page that used to reliably bring in visitors is now losing more than half of those clicks to the AI-generated answer sitting above it.
ChatGPT alone has 900 million weekly users and now handles roughly 12 percent of the volume Google search does.
And AI-driven traffic to websites has grown nearly 10 times over in the past year.
Here’s the part I actually find reassuring, though. Ahrefs shared their own data showing that in June 2025, AI search made up just half a percent of their total traffic, but it accounted for over 12 percent of their signups, a conversion rate 23 times higher than their organic traffic. When someone finds you through an AI answer, they’ve usually already had their question answered and are deciding who to trust, not just browsing. That’s a warmer lead walking through your door.
AEO Doesn’t Replace SEO, It Builds On It
If you’ve spent any time getting your website’s SEO right, breathe easy. None of that work is wasted. Good SEO, like solid content, real expertise, technical health, is still the foundation everything else sits on.
AEO is what you layer on top of your SEO foundation so AI tools can actually read, trust, and cite what you’ve built.
Sam Oh made a comparison in the course that stuck with me. Back around 2010, everyone said mobile apps were going to kill the web. Only old people would still use websites, they said. Apps did explode. But the web didn’t die, it grew right alongside them. AI search is likely following the same pattern. It’s growing fast, but Google still processes billions of searches a day, and traditional organic traffic still drives real business. The move isn’t to panic and abandon what’s working. It’s to play both games at once.
What This Actually Means for Your Business
This is where I want to bring it back down to earth, because “optimize for AI” can sound abstract until you break it into steps that actually fit into a small business’s week.
Find out where you stand first. Before you change anything, you need to know what AI tools currently see when they look at your site. I conduct AI Visibility Audits for exactly this reason, it’s the same process that took my own site from a visibility score of 57 to a 92, and took a Fort Worth marriage coach’s site from a 48 to a 90 using the same content, no redesign required.
You can’t fix what you haven’t measured.
Write new content with AEO in mind from the start. Going forward, every new page or post you publish can be built to answer the specific questions your future clients are actually asking AI tools, not just the keywords you’d type into Google five years ago. That’s a shift in structure and clarity more than a shift in effort.
Give your old content a second life. Most businesses I work with already have years of blog posts, service pages, and FAQs sitting on their site. Rather than starting over, the smarter move is usually rewriting and restructuring what’s already there so AI tools can actually find and trust it. That old content isn’t dead weight, it’s a head start most of your competitors don’t have.
You’re Probably Closer Than You Think
If you’ve already got decent SEO fundamentals in place, you’re not behind, you’re ahead of most of your competitors who haven’t even started thinking about this yet. The businesses that are going to struggle aren’t the ones who are new to AEO. They’re the ones who decide not to look at all.
You didn’t start your business to become an SEO or AI expert, and you shouldn’t have to. That’s what I’m here for. If you’re curious where your own site stands right now, an AI Visibility Audit is the easiest place to start, no pressure, no long-term contract, just a clear picture of what’s working and what to fix first.
Sources: This post draws on the first lesson of Ahrefs’ Answer Engine Optimization course, taught by Sam Oh. Watch it here: https://youtu.be/MLKgbeDeCxU. Course and data credit: Ahrefs (https://ahrefs.com/).
Site: LynnBusch.com — originally built in 2010, redesigned at least five times
Starting AI Visibility Score: 48/100 — Grade D
Ending AI Visibility Score: 90/100 — Grade A
What changed: missing schema, no way for AI tools to read the site, a duplicated headline, no FAQ content marked up for AI, and a server response time too slow to make Google (and ChatGPT) wait around.
Time to fix: one focused audit, worked through page by page and setting by setting
I Gave a Client’s Website an AI Visibility “Check-Up” (and the Diagnosis Wasn’t Pretty)
Lynn Busch is a Marriage Coach in Fort Worth. Her whole practice is built on noticing what’s actually going on underneath the surface — the pattern nobody named out loud, the thing a couple has been avoiding for years. So when I ran an AI visibility audit on her website, I liked the symmetry: I was doing for her site exactly what she does in her office. Looking past what looks fine on the surface to find what’s actually going on underneath.
What I found was a website that looked professional and put-together — and was, underneath, almost invisible to the AI tools that are increasingly how people find a therapist in the first place.
The AI Visibility “Exam”
If a client’s website were a patient, this is the medical intake exam I ran: could AI tools like ChatGPT, Claude, and Perplexity actually find the site, read it, understand what Lynn does, and trust it enough to recommend her? I used an AI Visibility Score tool to get a baseline, then went through the site page by page — homepage, specialty pages on infidelity and medical marriages, the works.
AI Visibility starting score: 48 out of 100. Not a website that was broken. A website that was quietly underperforming in ways nobody could see just by looking at it.
Here’s what the “exam” turned up:
No medical chart on file. No llms.txt — the file that tells AI tools what your site is and how to read it, the same way a new patient’s chart tells a doctor what they’re walking into.
Vitals running slow. Homepage response time (TTFB) was sitting at 836 milliseconds — sluggish enough to matter to both visitors and the crawlers trying to index the site.
No paperwork on the practice itself. No structured data (schema) identifying Lynn as a person, a local business, or a professional service — meaning AI tools had no reliable way to confirm who she is, where she practices, or what she specializes in.
A duplicated diagnosis. Two H1 headings effectively competing on the homepage — a leftover of how the page’s blurb modules had been built out over time — confusing AI crawlers about what the page is actually about.
Symptoms nobody had named. No FAQ schema on specialty pages, so genuinely useful answers to real questions (about infidelity recovery, medical marriages, premarital coaching) existed on the page but weren’t marked up in a way AI tools could pull out and quote.
No referral network on paper. Lynn’s social profiles (Facebook, Instagram, LinkedIn, Pinterest) weren’t connected back to the site through schema, so AI tools had no way to cross-reference and confirm she’s a real, established practitioner.
None of this was visible to someone landing on the homepage. All of it mattered to whether an AI tool would ever recommend Lynn Busch to someone quietly searching for help.
The AI Visibility “Treatment Plan”
Here’s what actually got treated, one issue at a time:
Ran labs and got the paperwork in order. Deployed custom JSON-LD schema — Person, LocalBusiness, ProfessionalService, and FAQPage types — through Divi’s Theme Options, giving AI tools a reliable, structured answer to “who is this, and what do they do.”
Filed the chart. Generated a curated llms.txt through the VigIA plugin (after disabling Yoast’s generic auto-generated version), so AI tools reading the site get an accurate, intentional summary instead of a default one.
Corrected the diagnosis. Fixed the duplicated H1 issue by converting stray H4 blurb modules down to proper H3s in the Divi visual builder, so the homepage has one clear, unambiguous headline again.
Connected the referral network. Populated the schema’s sameAs array with all four of Lynn’s social profiles, added her full practice address, and added a founding date — giving AI tools the cross-references they need to trust the practice is real and established.
Ran new vitals. Configured SiteGround Speed Optimizer with medium image compression and lazy load enabled. Homepage response time dropped from 836 milliseconds to 10 milliseconds.
Set the primary complaint. Set the homepage’s Yoast focus keyphrase to “relationship coach Fort Worth,” with a proper SEO title, meta description, and featured image — so both search engines and AI tools have a clear, consistent read on what the practice is and where it serves.
Added tools for ongoing monitoring. Installed VigIA, AI Content Signals, and AI Share & Summarize, so Lynn’s site now has ongoing visibility into how AI tools are seeing (and citing) it — instead of finding out a year from now that something quietly broke again.
Results of the AI Visibility Optimization Work
Score progression over the audit:
Initial audit …………………………… 48/100
After first pass …………………………. 67.4/100
After schema/plugin work ………………….. 75.4/100
After llms.txt + heading fixes ……………… 82/100
After speed/schema refinement ………………. 85/100
Final (settled overnight) ………………….. 90/100 — A+
Before and after:
Overall AI Visibility Score: 48/100 -> 90/100
Homepage response time (TTFB): 836ms -> 10ms
Same site, same content, same practice. No redesign. Just an exam nobody had run before, and a treatment plan worked through one finding at a time.
What I learned running this AI Audit
Nothing in Lynn’s site was unusual. It was the ordinary, invisible residue of a well-built website that had simply never had this specific kind of exam — because until recently, nobody was running one. That’s not a knock on Lynn’s site, or on the developer who built it. It’s just what happens to every website right now: the rules for being found changed faster than most site owners had a reason to notice.
This is now the audit I run for clients who’ve had that nagging feeling — that their site should be doing better than it is, that people who’d be a great fit somehow aren’t finding them. If that’s you, let’s talk about running an AI Visibility Audit on your site: https://sosmymarketing.com/contact/
Retrieval Augmented Generation (RAG) is the technology that lets AI tools like ChatGPT, Google AI Overviews, and Perplexity search the web or a trusted database for current, relevant information before writing an answer — instead of relying only on what they memorized during training. For small businesses, this matters because RAG is how AI chatbots decide which websites to read, trust, and quote when a potential customer asks a question. If your website content is clear, well-organized, and easy for AI to find and understand, RAG-powered tools are more likely to pull your business into their answers — putting you in front of customers before they ever click a traditional search result.
Picture two students taking the same test.
The first student has to answer from memory alone — everything they know is whatever they studied months ago. If something changed since then, or they never learned it, they either guess or get it wrong. That’s how older AI chatbots worked: they generated answers purely from patterns learned during training, which is why they sometimes made things up or gave outdated information.
The second student gets to take an open-book exam. Before answering, they flip to the right page, find the current facts, and then write their answer using what they just looked up. That’s RAG.
In practice, RAG works in two simple steps:
Retrieve — When someone asks a question, the AI first searches an index of web pages, documents, or a database to find the most relevant, up-to-date information.
Generate — The AI then reads what it found and writes a natural-sounding answer based on those real sources, often citing or linking back to them.
This is the technology quietly running behind LLMs like Google’s AI Overviews, ChatGPT’s web search, Microsoft Copilot, and Claud. Every time one of these tools answers a question with current facts and a source link, RAG is doing the work behind the scenes.
Why RAG Matters for Marketing Your Business
Here’s the part that should get your attention as a business owner: RAG determines who gets mentioned in AI-generated answers, and who gets left out.
Think about how people search today. A potential customer doesn’t just type a keyword and scroll through ten blue links anymore — they’re increasingly asking an AI assistant a full question, like “What’s the best marketing agency for a small bakery in my area?” or “How do I get more Google reviews?” The AI reads across many websites, retrieves the most relevant and trustworthy ones, and stitches together an answer — sometimes naming specific businesses along the way.
If your website is never retrieved, you’re invisible in that conversation. It doesn’t matter how great your product or service is if the AI never “sees” your content to consider it. On the flip side, if your business does get pulled in and cited by name, you get something even more valuable than a click: you show up as the trusted, recommended answer, in front of a customer at the exact moment they’re deciding who to hire or buy from.
In short, RAG has turned “being findable” into a marketing requirement, not just a technical nicety.
Why RAG Means SEO Still Matters
A lot of business owners hear about AI-generated answers and assume SEO is dead. It’s actually the opposite — RAG makes good SEO more important, not less.
Here’s why: RAG systems don’t retrieve information out of thin air. They pull from search indexes and crawled web content — many of them built on the same core ranking systems and quality signals that traditional SEO has always optimized for. An AI tool still needs your site to be:
Crawlable and indexable — if search engines can’t find and read your pages, neither can the AI retrieving from them.
Authoritative and trustworthy — AI systems lean heavily on trust signals (real expertise, credible reviews, consistent business information) to decide which sources are safe to quote.
Well-structured and current — clean headings, clear organization, and up-to-date facts make it far easier for retrieval systems to find and pull the right piece of your content.
Think of it this way: SEO gets you discovered and trusted enough to be in the running. RAG (and the newer discipline growing alongside it, often called Answer Engine Optimization or AEO) determines whether you actually get chosen and quoted once you’re in that running. They work together, not against each other. Businesses that skip SEO fundamentals won’t just fall in Google rankings — they’ll become invisible to AI tools too.
Tips to Optimize Your Website Content for RAG
The good news is that most RAG-friendly practices are extensions of good content habits, not a whole new playbook. Here’s where to focus:
Answer questions directly and early. Open each page or section with a clear, concise answer before adding supporting detail. RAG systems (and impatient readers) reward directness.
Write in natural, question-based headers. Structure content around the real questions your customers ask — “How much does X cost?” or “How long does Y take?” — using those phrases as headings. This mirrors how people actually query AI tools.
Keep each section focused on one idea. RAG systems break pages into small “chunks” of text and retrieve the most relevant chunk, not the whole page. A section that mixes five topics together is much harder to retrieve accurately than one tightly focused paragraph.
Add structured data (schema markup). FAQ, Article, LocalBusiness, and Review schema help machines understand exactly what your content means, not just what it says.
Keep facts current. pdate pricing, hours, service areas, and offers regularly. RAG systems favor fresh, accurate information over stale pages.
Build real authority signals. Author bios, genuine customer reviews, case studies, and consistent business details across the web all help establish the trustworthiness AI systems look for.
Cut the fluff. ong-winded introductions and vague marketing language make it harder for both AI and human readers to find your actual answer. Clear, plain writing wins.
Make sure your site is technically sound A current sitemap, no accidental blocks in your robots.txt file, and fast page load times ensure your content can be crawled and retrieved in the first place.
RAG isn’t a reason to abandon your marketing fundamentals — it’s a reason to sharpen them.
The businesses that show up in AI-generated answers tomorrow will be the ones with clear, well-structured, trustworthy content today. If you’d like help auditing your website and content to make sure it’s ready for both search engines and AI, that’s exactly what we do at SOS My Marketing.
Site: sosmymarketing.com — originally built in 2008, redesigned more times than I can count
Starting AI Visibility Score1: 57/100
Ending AI Visibility Score: 92/100 — Grade A+
What changed: broken links, missing image labels, no way for AI tools to read the site, a duplicated headline, and hosting settings that had been quietly slowing everything down for years
Time to fix: one focused audit, worked through page by page and setting by setting
My website, sosmymarketing.com has been online since I started my business in 2008. In internet years, that’s ancient. Over nearly two decades, this site has been redesigned, re-platformed, re-templated, and redecorated more times than I can honestly count. Every redesign added something new. Almost none of them cleaned up what came before.
Sound familiar? If you’ve had a WordPress site for more than a few years, it probably does.
I finally did what I tell my clients to do all the time: I stopped adding new rooms and went down to the basement with a flashlight. What I found was a website that looked fine from the street but had years of junk piled up where visitors — and increasingly, AI tools like ChatGPT, Claude, and Gemini — couldn’t see it. So I ran a full AI visibility audit, cleaned house, and want to walk you through exactly what that looked like, because if your WordPress site has a similar history, you’re probably sitting on the same kind of mess.
The House Nobody Inspected
This isn’t a metaphor I’m reaching for. Back in 2008, when I started this business, I taught classes for entrepreneurs on how to build their first website — and I explained it exactly this way. The URL is the land. Hosting is where you put the bones of the house. WordPress is how you decorate that house with words and pictures. Ah, how simple things were back then.
Eighteen years later, it turns out the same analogy is the most honest way to describe what happens to a house nobody’s fully inspected since it was built. Think of a website like a house you’ve owned for many years. You’ve repainted, you’ve swapped out the furniture, maybe you knocked down a wall or two. But nobody’s been in the attic since 2012, and nobody’s checked what the last three contractors left behind in the basement.
That’s what an old WordPress site usually looks like under the hood:
Doors that don’t open. Broken internal links and redirect chains — pages pointing to pages that pointed to pages that don’t exist anymore.
Unlabeled rooms. Images with no alt text, meaning neither search engines nor AI tools have any idea what’s actually in them.
No address on file. No llms.txt or llms-full.txt — the modern equivalent of AI tools not having your address, so they can’t reliably find or understand your site.
A duplicated front door. Two H1 headings on the homepage, which confuses search engines and AI crawlers about what the page is actually about.
Old plumbing. A slow server response time (TTFB) from caching that had quietly been half-broken for who knows how long.
None of this is visible to a regular visitor scrolling through the homepage. All of it matters enormously to the search engines and AI systems trying to understand, index, and recommend your site.
The Inspection: Running the Numbers
I used an AI Visibility Score tool that grades a site across five categories — think of it as a home inspector’s checklist for whether AI tools can actually find, read, and trust your site:
Category
What it really means
Starting score
Access and AI discovery
Can AI crawlers even get in the front door?
20/37
Structured data and semantic context
Does the site have its paperwork in order — the deed that says what it actually is?
16/25
Content structure and readability
Is the house laid out logically, with rooms that are clearly labeled?
13/20
AI interaction and distribution
Can visitors (including AI tools) actually get around once they’re inside?
4/8
Access performance
How fast does the water actually come out of the tap?
4/10
Total: 57 out of 100. Not falling down. Not exactly move-in ready either.
The Renovation
Here’s what actually got fixed, room by room:
Fixed the doors. Cleaned up broken redirect chains and 404s across the site, and updated internal links that were still pointing at old, redirected URLs.
Labeled every room. Used AI to write proper alt text for every image across the site — 27 out of 66 images were missing alt text completely — so both screen readers and AI tools know exactly what they’re looking at, not just that “an image exists here.”
Filed the paperwork. Verified and cleaned up the structured data (schema) that tells search engines and AI tools what kind of business this actually is, not just what the words on the page say.
Put the address on file. Generated llms-full.txt — a complete, AI-readable copy of the site’s content — and turned on Markdown endpoints, so any AI agent that wants to read a page cleanly, without wading through a decade of design layers, now can.
Fixed the duplicated front door. Found two H1 headings stacked on top of each other on the homepage (leftover from an old responsive design pattern) and cleaned it down to one, like it should have been all along.
Rewired the electrical. This was the big one. The site’s hosting had three layers of caching available, and only two were switched on — the object cache (Memcached) had silently been off. Once it was properly connected, homepage response time (TTFB) dropped from 701 milliseconds to 22 milliseconds. That’s not a typo. That’s the difference between a slow drip and a fully open tap.
Checked who’s allowed on the property. Also found a years-old list of entire countries blocked at the hosting level — including the UK, Canada, Germany, and Singapore. Nobody remembered why it was there. Beyond turning away legitimate visitors from major markets, it was likely blocking AI crawler traffic too, since Germany and Singapore both host major cloud data centers. Trimmed it down to what actually made sense to keep.
Results of the AI Visibility Optimization Work
Category
Before
After
Overall
57/100
92/100 — A+
Access and AI discovery
20/37
37/37
Structured data and semantic context
16/25
21/25
Content structure and readability
13/20
16/20
AI interaction and distribution
4/8
8/8
Access performance
4/10
10/10
2Same site. Same content, mostly. No redesign. Just years of accumulated junk cleared out, room by room, until search engines and AI tools alike could actually walk through the whole house.
What I Learned Cleaning My Old Website
The honest truth: nothing I found was exotic. It was the ordinary residue of a site that had been through a lot of hands and a lot of years without anyone doing a full inspection. That’s not a special SOS problem — it’s a WordPress problem. WordPress makes it easy to keep adding rooms. It doesn’t do much to make you clean out the ones you’re not using anymore.
If your WordPress website has been through a few redesigns and nobody’s checked what’s in the attic, I can help. That’s exactly the kind of audit I just ran on my own site — and it’s the same process I use for clients who’ve had the same nagging feeling that their site “should be doing better than it is.”
P.S. A few days after finishing the renovation, Google Search Console flagged one more thing: an old, nonexistent URL — a literal wildcard character where a page should be — that it had crawled and never found. I traced it back through the sitemap, robots.txt, redirects, and templates expecting to find one more leftover mess. Instead, I found nothing. It had already been fixed, quietly, as a side effect of all that redirect cleanup during the renovation. Sometimes you find out you patched the leak before you even knew it was there.
