AI visibility is the difference between being the brand a buyer is told to consider, and being the brand the model never mentions. Traditional SEO still decides whether you can be found. AI visibility decides whether you are named once the answer is written for the user.
This is not a thought piece about “the death of search”. Google remains the default starting point for most queries. What has changed is what happens after the query. In the first four months of 2026, SparkToro and Similarweb found that 68.01% of US Google searches ended without a click (SparkToro / Similarweb, June 2026). That is up from 60.45% in 2024, and from about 49% when SparkToro first measured the “zero-click web” in 2019. The open web is still there. A growing share of demand never leaves the results page, the chatbot, or the overview.
If you run SEO for an agency, an in-house team, or your own company, you now need two scoreboards: where you rank, and where you are cited. This article explains the evidence, the vocabulary (AEO, GEO, AI visibility), a measurement workflow you can run this month, and how Odyssey Advance puts those jobs in one platform instead of a pile of browser tabs.
Marketers have been quoting “AI will kill SEO” since early 2024, often from a single Gartner forecast: that traditional search engine volume would drop 25% by 2026 as chatbots absorbed queries (Gartner, 19 February 2024). That line was useful as a provocation. It has been a poor description of what happened next. Aggregate search demand did not collapse. Google absorbed the behaviour change by putting the answer on the results page. Click-through rates, not query volume, took the hit.
Three studies are worth putting on the same slide, because they measure different things and still point the same way.
Pew Research Center’s work is the cleanest behavioural study in the set. Researchers analysed browsing from 900 US adults in March 2025: 68,879 unique Google searches, of which 12,593 produced an AI-generated summary. On pages with a summary, users clicked a classic blue link in 8% of visits, against 15% without one. Clicks on the citations inside the summary itself happened in about 1% of visits. Sessions were also more likely to end entirely when a summary appeared (26% versus 16%) (Pew Research Center, 22 July 2025).
Seer Interactive then showed what that looks like in Search Console over time. Across 3,119 informational queries, 42 organisations, and 25.1 million organic impressions from June 2024 to September 2025, organic CTR on queries with AI Overviews fell from 1.76% to 0.61%, a 61% decline. Paid CTR on the same set fell 68%, from 19.70% to 6.34%. Brands that were cited inside the overview still suffered, but less: they earned roughly 35% more organic clicks and 91% more paid clicks than uncited brands on comparable queries (Seer Interactive, 4 November 2025). Seer is clear that this is an association, not proof that citation causes the extra clicks. Stronger brands may simply be more likely to be cited. Either way, citation sits next to a residual click advantage in a thinner click pool.
Semrush’s 2025 study on more than 500 digital marketing and SEO topics adds the commercial twist. Visitors who arrived from non-Google AI sources (ChatGPT and similar) converted at 4.4 times the rate of traditional organic visitors. Volume is still small. Quality is not. The same research found ChatGPT weekly active users had grown eight-fold from October 2023 to April 2025, to more than 800 million, and projected that for those topics AI search visitors could overtake traditional search visitors around early 2028, sooner if Google’s AI Mode becomes the default results page (Semrush, July 2025).
Put simply: fewer people click. The people who do click after an AI answer are often further down the decision. And a large share of influence now happens in a paragraph you do not control, on a page you do not own.
If your only KPI is organic sessions, you will report a decline and miss the moment your category started being decided in the answer.
AI visibility is the measure of how your brand, products, people and pages appear inside machine-generated answers. That includes Google AI Overviews and AI Mode, ChatGPT, Claude, Perplexity, Gemini, Copilot, and the long tail of assistants that will sit inside browsers, operating systems and ecommerce search.
It is not one score. A useful programme tracks at least five things, because a single “share of voice” number hides the failure modes:
A brand can “win” a mention and still lose the sale. If ChatGPT says you are a good option for enterprises when you sell to UK SMEs, you have visibility and a positioning problem. If Perplexity cites a 2022 blog post you meant to retire, you have a freshness problem. If Google’s overview names a review site that misstates your pricing, you have a third-party evidence problem. None of that shows up in a keyword rank tracker on its own, which is why rank tracking and AI visibility have to live together.
AI visibility is also not “stop doing SEO”. Semrush found that when ChatGPT search cites pages, those pages sit in organic positions 21 or lower almost 90% of the time. That sounds like traditional rankings are irrelevant. The same study noted that some models still prefer pages that already rank in positions 1 to 5 more often than those in 6 to 20 (Semrush, July 2025). Crawlability, entity clarity, and useful pages still feed both systems. The change is that a page 24 result with a clean answer can be cited, while a page 1 result that is thin, JavaScript-only, or off-intent can be ignored.
The industry has minted acronyms faster than it has minted dashboards. You will hear AEO (answer engine optimisation) and GEO (generative engine optimisation) used as if they replaced SEO. They did not. They describe additional surfaces. The work overlaps. The measurement does not.
| Practice | Primary question | Typical evidence | What “winning” looks like |
|---|---|---|---|
| SEO | Can Google find, understand and rank this page? | Rankings, impressions, clicks, coverage, links | You appear for the query, preferably with a SERP feature |
| AEO | Can an answer engine extract a correct, citable span? | Featured snippets, AI Overviews, People Also Ask, citations | Your facts are quoted; your brand is the source |
| GEO / AI visibility | Do generative models recommend you in a conversation? | Prompt sets, mentions, sentiment, competitors named | You are in the shortlist the buyer hears |
| Technical SEO | Can crawlers and models fetch the real content? | Indexation, canonicals, robots, JS rendering, llms.txt | The right URL is the one that gets read |
Use the labels internally if they help stakeholders. Report the evidence, not the acronym.
Nobody outside the labs has the full ranking function, and anyone selling you a guaranteed “GEO hack” is selling incense. We do have consistent patterns from public studies and from watching live answers.
Third-party proof still punches above owned media. Semrush’s topic study found Quora was the most-cited website in Google AI Overviews, with Reddit second. The same research found that 50% of links in ChatGPT 4o responses pointed to business or service websites, even though individual forum domains perform strongly (Semrush, July 2025). That does not mean you should abandon your site for Reddit threads. It means a page that only says “we are the leading provider” with no numbers, no definitions, and no comparison is easy for a model to skip.
Chunkable structure wins over atmosphere. Models lift spans: definitions, steps, tables, FAQs, specification rows. A 2,000-word essay with the answer buried in paragraph 14 is harder to quote than a heading that asks the question and a paragraph that answers it in the first two sentences. This is why on-page content optimisation still matters: not as a density ritual, but as a way to make the extractable unit obvious.
Entities have to be boringly consistent. If your homepage, LinkedIn, directory listings and press mentions disagree on the company name, product names, locations or category, the model has to guess. Guessing is how you become “a marketing tool based in London” when you are a Birmingham platform. Schema, about pages, and a factual company page are not brand fluff in this environment. They are disambiguation.
Crawlability is still a gate. If important copy only appears after client-side JavaScript, many AI crawlers will not see it. If the canonical points at a thin duplicate, you may train the model on the wrong URL. If the page is noindexed, you should not be surprised when nothing cites it. Pair AI visibility work with an SEO site auditor and an indexation checker. You cannot be recommended from a URL the machines cannot fetch.
The prompt is the new keyword, but it is not a synonym list. Buyers do not type “best CRM software UK 2026 pricing comparison for 40-seat professional services firms” into Google every day. They do type versions of that into ChatGPT. Your research set has to include:
That is why keyword research is still the start of the work. The seed terms tell you the language. The prompt set tells you the conversation. Odyssey Advance is built so those two views are not trapped in different products.
You do not need a 40-prompt science project in week one. You need a set you will actually re-run. For most UK SMEs and agencies, 25 to 40 prompts per product line is enough to see the shape of the problem.
Odyssey Advance combines AI visibility tracking with keyword research, rank tracking and technical audits in one workspace. Create an account or book a walkthrough on your own site.
View pricing Book a demoWeek one is inventory. Export your top landing pages, your converting keywords, and the five competitors sales actually lose to, not the five you wish you competed with. Write 30 prompts. Include at least five that a sceptical buyer would ask (“alternatives to…”, “is it worth it”, “cheapest”, “for small teams”).
Week two is diagnosis. Run the prompts. You are looking for three piles: prompts where you are absent, prompts where you are present but misdescribed, and prompts where a competitor owns the recommendation. Open the cited URLs. If the model is learning you from a partner page, a G2-style listing, or a stale blog, that is your content calendar, not a mystery.
Week three is the unglamorous repair. Fix indexation and canonical issues on the URLs you need cited. Add a comparison section that states differences without insulting the category. Put numbers, definitions and FAQs at the top of service pages, not below a hero that says “reimagine”. Use the keyword density tool as a check against stuffing, not as a target. Models are not impressed by a 3.2% density. Humans are put off by it.
Week four is proof and process. Retest. Record what moved. Give whoever owns PR a list of the domains that already get cited in your category. Give whoever owns content a list of questions that still have no citable paragraph on your site. Put the scorecard in the monthly report next to rankings, not in a separate “innovation” appendix that nobody reads.
None of this requires you to believe that ChatGPT will replace Google next Tuesday. Pew, SparkToro and Seer are already describing the present: fewer clicks, more on-SERP answers, a measurable bonus for being cited, and a second discovery channel whose visitors convert unusually well when they do arrive.
Most teams try to cover this with a rank tracker, a crawler, a ChatGPT Plus window, and a junior staff member pasting answers into Notion. That lasts until the first holiday. Odyssey Advance exists because that workflow does not survive contact with a client roster.
In one suite you can research the terms that still carry demand, track where those terms sit in classic results (including snippets and local packs), crawl the site that is supposed to earn the citation, and measure how generative engines talk about the brand. Guided content then takes the same keyword set into a draft you can actually publish. That is the loop: model the demand, diagnose the answer, ship the page, retest.
You can read more about the product on our About Us page, or open an account.
If you only take one idea from this piece, take this: rankings tell you if you can be chosen. AI visibility tells you if you were. Run both, on a cadence, with the same people who already own SEO. The organisations that treat generative answers as a side project will spend 2027 explaining a traffic chart that no longer matches the pipeline.
When you are ready to ship pages that can actually be quoted, use the companion guide on writing content that ranks and gets cited. If the blocker is five logins rather than five ideas, read why marketing teams are consolidating SEO tools.
No. Semrush’s citation work suggests models often pull from pages that are not in the traditional top three. Niche, specific, well-structured pages can be cited. What smaller teams lack is a repeatable way to see it. That is a tooling problem, not a brand-size law.
Yes. CTR is how you notice the overview tax. It is a worse north-star than it was in 2019, because so much influence is now clickless. Pair CTR with branded search, assisted conversions, and AI mention rate.
Only if your SEO team refuses to own prompts, citations and third-party evidence. The skills overlap with content, technical SEO and digital PR. Splitting them into two retainers usually produces two dashboards and one confused client.
Start with keyword research and the rank tracker on the terms you already monetise, then layer AI visibility prompts on the same entities. Repair crawl and indexation issues before you scale content. Book a demo if you want that path walked on your domain.