Category: AI SEO

A founder asked me last week if he could “skip the SEO stuff” and just optimize for ChatGPT and Perplexity. He’d read three breathless posts about GEO, AEO, and citation engineering. He thought traditional SEO was the past.

Here’s what I told him, and what every recent dataset confirms: there is roughly a 92% correlation between pages that rank in the organic top 10 on Google and pages that get cited in Google AI Overviews. Your blue-link rankings are not a separate world from AI visibility. They are the on-ramp.

That is the unglamorous truth nobody wants to publish in a hot take.

The mechanic — why retrieval looks a lot like ranking

AI search engines don’t read the open web in real time. They use retrieval pipelines: candidate pools, relevance scoring, freshness filters, source-quality weighting. The candidate pool for ChatGPT, Perplexity, AI Overviews, and Gemini is built on the same signals classical search uses — crawlability, internal linking, topical depth, query-page relevance, authority.

If a page can’t be crawled cleanly, an LLM can’t ingest it. If it isn’t relevant for the query, it isn’t pulled into the candidate set. If it’s slow, the embedding pipeline deprioritizes it (pages with FCP under 0.4s average 6.7 citations versus 2.1 for over 1.13s). Bad SEO fundamentals are upstream of bad AI visibility. No amount of “GEO content” survives a broken crawl path.

Then the AI layer adds its own filters on top — answer-unit structure, statistic and quote density, entity consistency, fresh updates. But those filters operate on the candidate pool that classical SEO built. If you’re not in the pool, none of it matters.

What a “skip the basics” strategy actually looks like

I’ve audited a dozen sites this year that bought the “GEO is different” pitch. Same pattern every time: thin technical SEO, slow page speed, broken internal links, no entity consistency between site and Wikidata, no schema, and a flurry of new “ChatGPT-optimized” content sitting in folders Googlebot has never visited. Their AI visibility was zero. Of course it was — they were invisible to the layer underneath.

The fix is never spectacular. It’s the same boring list it’s been for fifteen years, with a few modern items added at the end:

• Crawlable site, clean canonical tags, no orphan pages
• Strict H1 → H2 → H3 hierarchy on every important page (68.7% of cited pages do this)
• Internal links from authoritative pages to deep ones
• Page speed under 1 second to FCP wherever possible
• Real backlinks from real sites — sites with 32K+ referring domains are 3.5× more likely to be cited by ChatGPT than sites with under 200
• Then layer on the new stuff — entity work, citation-engineered paragraphs, freshness cadence, schema

The new stuff is real. It just sits on top of the old stuff. Skip the foundation and you are decorating a building you haven’t poured.

Why the 92% number matters strategically

A 92% overlap between the organic top 10 and AI Overview citations means something specific: ranking work is dual-purpose. Every dollar spent making a page rank #6 is also a dollar spent making it citable. You don’t have to fund two parallel programs. You have to fund one program that ends in two outcomes.

That reframes the budget conversation in every agency-client call I’ve had this year. “GEO” isn’t a separate line item. It’s a layer added to the SEO line item. Anyone selling it as a parallel discipline is either confused or selling 2× the hours for 1× the work.

What to do this week

1. Pull your top-10 ranking list. Filter for any page on a high-intent commercial query. That is your AI-citation candidate pool. Fix indexability, page speed, internal links, and heading hierarchy on those pages first.

2. Compare your ranking pages to your AI-cited pages. Run a manual prompt audit — ask ChatGPT, Perplexity, AI Overviews, and Gemini your top 20 queries and log which URLs they cite. The overlap tells you where retrieval is doing its job. The non-overlap tells you which pages need on-page AI work — answer units, stats, quotes, entity reinforcement.

3. Audit any “GEO-only” content for traditional SEO basics. If a page can’t be reached in three internal clicks from your homepage, it isn’t in any candidate pool — AI or otherwise.

4. Stop writing new “GEO content” until the foundational pages are clean. New articles inherit the site’s authority signals. Pouring content onto a broken site is the slowest possible way to be cited.

The agencies pitching “GEO replaces SEO” are selling a story. Retrieval pipelines, candidate pools, and a 92% top-10 correlation are saying something quieter and more useful: do the unsexy work, then add the new layer.

If you’re a brand that wants to be the answer LLMs reach for (not just rank on Google), Paris Roussos has been engineering search visibility for 30 years and now runs done-for-you AI SEO. Flat-rate, no-fuss. Email parisroussos@gmail.com.

The future of AI SEO looks a lot like the past — only the people who kept doing the boring parts are getting cited.

If you’ve watched Search Console month over month for the last year and felt like someone unplugged a wire, you weren’t imagining it. The wire got unplugged. Ahrefs and Seer pegged the organic CTR drop on AI Overview queries at 58–61%. Paid is worse — about 68% down. A handful of publishers reported declines as deep as 89%. The single biggest change to organic in a decade happened quietly, and most marketing teams are still optimizing for the old math.

But there is a second number nobody is putting on a slide, and it’s the one that matters: brands cited inside AI Overviews are pulling +35% organic clicks and +91% paid clicks on the same queries. The ranking page is no longer the prize. The citation slot is.

What’s actually happening behind the scenes

AI Overviews don’t replace search — they intercept it. Google’s Gemini-powered layer reads the top-N retrieval set, summarizes it, and quotes a small handful of sources directly inside the answer card. Users get the answer. They scroll less. Most of them never reach the blue links. That’s where your 58% went.

The brands that come out ahead are the ones whose names, products, and exact phrases get pulled into the summary itself. When a user reads “according to [Your Brand], the answer is X,” two things happen: the brand earns trust without a click, and a meaningful slice of users does click — to verify, to buy, to read the source. Amsive’s data on cited brands isn’t a quirk. It’s the new shape of the funnel.

Here’s the part operators miss: the citation pool is not the top 10. It’s a different pool. Earlier work on AI Overviews showed 83% of citations come from outside the traditional top 10. AI Overviews aren’t ranking your homepage. They’re surfacing the deep, specific, well-structured page you forgot you wrote three years ago — if it answers the question cleanly.

What this means if you sell things

Stop measuring impressions on AIO queries. They’re noise now. Start measuring two things:

1. Citation share — for the 50–200 queries that actually drive your business, how often is your brand inside the AIO box? You can spot-check manually, but real visibility tracking (Profound, Peec, Ahrefs Brand Radar, or your own scraper) is now table stakes.

2. Branded query lift — when AIO mentions you, your branded search volume should rise within 7–14 days. If it doesn’t, your AIO mention isn’t sticky enough — usually because your brand name appears as bare text instead of next to a memorable claim or stat.

The CTR-loss panic is the wrong panic. The real fire is that competitors who get cited are quietly cannibalizing your unbranded demand and feeding their own branded demand. That gap compounds.

What to do this week

You don’t need a quarter-long GEO program to start moving the needle. Four moves, ranked by ROI:

• Pick your 20 highest-revenue queries and check who’s getting cited in AIO. Not “who’s ranking” — who’s quoted. Note the cited site, the exact sentence pulled, and the source URL. Patterns will jump out within an hour.
• Rewrite the top of your money pages so the strongest, most quotable claim is the first 40–60 words under the H1. AI Overviews extract from the top of the page disproportionately. Bury your hook and you forfeit the slot.
• Audit which of your pages are deep enough to be cited. A 600-word product page won’t survive against a 2,000-word competitor page that names entities, cites stats, and uses strict H1→H2→H3 structure. Pick three pages this week, expand them, and add a stat-and-quote pair to each — that combination alone correlates with a measurable visibility lift.
• Set up a weekly citation check. A simple spreadsheet — query, AIO citation Y/N, who got it, what they said — tells you in three weeks whether your moves are working. Without measurement, you’re guessing.

The teams winning this cycle aren’t necessarily writing more. They’re writing the same volume, but every page is built to be quoted.

Need this done for you? Paris Roussos runs a flat-rate AI SEO service ($500–$1,500/mo per client, white-label for agencies) covering audits, schema and entity work, AI-visibility tracking, and content engineered to be cited by LLMs. Reach him at parisroussos@gmail.com.

Optimize for the citation, not the click — the citation is what brings the click back.

If you have been writing for human readers for the last twenty years, you have been trained to keep your prose clean. Drop the jargon. Cut the data dump. Tell the story. That instinct is now actively hurting your AI visibility.

LLMs are not skimming for narrative. They are extracting answer units — discrete, citeable pieces of information they can lift into a response and attribute. The two units they lift most often are statistics and direct quotations. If your page is missing both, you have given the model nothing to reach for.

The mechanic: extraction, not interpretation

Across the 2025–2026 GEO benchmark studies, two numbers keep showing up. Pages with embedded statistics see roughly 22% more AI visibility. Pages with direct quotations from a named source see roughly 37% more. Stack both on the same page and the lift compounds — not because the LLM “likes” data, but because data is what the retrieval and synthesis pipeline is built to grab.

When ChatGPT or Perplexity assembles an answer, it is not reading your post the way a human does. It is breaking the page into chunks, embedding each chunk, retrieving the top-N most relevant chunks for the user’s query, and then asking the model to draft a response grounded in those chunks. A chunk that contains a number with a source (“44.2% of citations come from the first 30%”) or a quoted human expert (“‘AIO has crushed CTR but rewarded cited brands,’ said the Ahrefs team”) is a high-confidence chunk. The model can ground a sentence in it without paraphrasing risk. It will be picked over your beautifully written third paragraph nine times out of ten.

Google’s AI Overviews work on the same principle. So does Gemini. So does Claude when it is doing search. The chunk-level extraction logic is now industry-standard, and the chunks that get picked share a profile: a clean assertion, a number or quote, and an attribution.

What the average page looks like (and why it loses)

Pull up your top-traffic blog post from 2023. Count the statistics. Count the named quotes. If you find one stat in the intro and zero quotes, you have a typical page — and a page the LLMs will skim past in favor of an Ahrefs blog post or a Yext write-up that loaded the same topic with five stats and three named experts.

The fix is not to turn your site into a research paper. It is to seed each major section with one citeable unit. A section on AI Overviews CTR loss should contain the actual percentage (58–61% organic CTR drop, per the Seer / Ahrefs studies). A section on schema should contain a researcher quote or a platform statement. A section on freshness should name the 2-month / 28% lift number. One unit per section. That’s it.

What to do this week

Open the four or five pages on your site that you actually want LLMs to cite — your money pages, not your archive. For each one, do this:

1. Add at least three statistics with sources. Not vague ones (“most marketers say…”). Specific, sourced, dated. “44.2% of LLM citations come from the first 30% of a page (ConvertMate, 2026).” If you don’t have your own data, pull from published research and link it. The link itself is a signal.

2. Add at least one direct quote per major section. A real quote from a named person at a named company. Industry analysts, your own founder, a client testimonial — all qualify, as long as they are attributed. Use real quotation marks, not paraphrase. Models extract quotes by punctuation pattern.

3. Front-load the heaviest one. If you only do one thing, put a big stat in the first 100 words of the page. The first 30% of the page produces the bulk of citations — a stat there is doing double duty.

4. Re-verify quarterly. A stale stat from 2022 hurts you. Refresh sources every quarter, update the year in the citation, and let the freshness signal compound with the citation signal. (See last week’s post on the 2-month refresh rule.)

This is one of the highest-ROI moves in the AI SEO playbook right now because the ceiling is real (a +22% to +37% lift, often more in combination) and the work is mechanical. You are not rewriting strategy. You are bolting citeable units onto pages you already published.

Need this done for you? Paris Roussos runs a flat-rate AI SEO service ($500–$1,500/mo per client, white-label for agencies) covering audits, schema and entity work, AI-visibility tracking, and content engineered to be cited by LLMs. Reach him at parisroussos@gmail.com.

The brands getting cited in 2026 aren’t the best writers. They’re the ones who made their pages easiest to lift.

If your evergreen library hasn’t been touched in a year, AI search is quietly skipping past it.

Server-log studies across 2025 and 2026 show that 65% of AI bot crawl activity targets pages published in the past 12 months, and pages updated within the last 2 months earn 28% more citations than older pages. ChatGPT, Perplexity, Google AI Overviews and Gemini are not running the same retrieval logic Google search ran in 2018. They lean toward fresh — sometimes aggressively — because their training is anchored in a fixed date and their live retrieval layer is the only place they can pick up “what’s true now.”

That has practical consequences for any agency or operator who built a content library on the assumption that an evergreen post written in 2022 still pulls weight. It does in classic organic. It does not in answer engines.

The mechanic

LLM-driven search splits into two phases. First, the model has its training cutoff — a frozen snapshot of the web. Second, when a user asks a current question, the system retrieves live pages from a search index and uses those as grounding sources for the answer.

The retrieval layer is where freshness wins. The retrieval index is biased toward recently published or recently updated URLs because (a) crawlers reprioritize pages with new lastmod dates, (b) embedding pipelines reweight chunks tied to current entities and timestamps, and (c) ranking models inside the retrieval stack treat staleness as a soft negative signal — the older the page, the more likely it has been superseded.

Add to this that LLMs prefer to cite sources that look authoritative right now. A page dated 2021 introducing GA4 reads differently than a page updated last month covering the same ground with current screenshots and current numbers. The retrieval layer sees both. The citation usually goes to the second.

This is also why “publish and forget” content programs underperform. A 200-article library with 6 fresh pieces a year will lose visibility to a 50-article library refreshed continuously.

What to do this week

Pick the 20 pages on your site (or your client’s site) that earn the most organic traffic and should be earning AI citations. For each one:

1. Update the lastmod date — but only after a real edit. Stamping a page without changing the body is a short-lived trick and AI retrieval systems are already discounting it. Add a fresh stat, a new example, a 2026 link.

2. Change the intro. 44.2% of LLM citations come from the first 30% of the page. If the lede still references “in the post-COVID landscape” or any 2022-flavored framing, rewrite it.

3. Add or refresh dated proof. Replace 2023 numbers with 2025/2026 numbers. Cite a recent named study with its publication date in the sentence — LLMs love that pattern because it makes provenance clean.

4. Touch the page every 60 days going forward. Calendar it. The 2-month window is the operational sweet spot the data supports.

5. Rebuild your sitemap weekly. Refreshed pages need a re-crawl trigger or the freshness signal sits unused.

For agencies running multi-client portfolios, this is the easiest “look how AI SEO is different” demo you can run. Pick a high-value page, refresh it, watch the citations pick up over the following month, screenshot the lift. Clients understand that loop instantly.

The deeper point is that AI search rewards maintenance in a way classic SEO did not. Classic SEO let you ship a comprehensive guide and ride it for years. Answer engines pay you for being the most current source on the question. The work is no longer front-loaded; it’s continuous.

The agencies that are going to win the next 24 months of this shift are the ones who quietly fold a refresh cadence into every retainer — not the ones still pitching one-and-done content sprints.

---

If you’re a brand that wants to be the answer LLMs reach for (not just rank on Google), Paris Roussos has been engineering search visibility for 30 years and now runs done-for-you AI SEO. Flat-rate, no-fuss. Email parisroussos@gmail.com.

Refresh isn’t a chore. It’s the lever.