How can I improve both my SEO and AI visibility?
Running these as separate programmes is the most common and most expensive mistake in the field. Most of the work is shared; the trick is knowing which part is not.
The most expensive mistake in this field is running SEO and AI visibility as two programmes. They compete for the same engineering time, make contradictory edits to the same pages, and produce two reports that never reconcile.
They are one programme with a divergence at the edges. Knowing exactly where the divergence is makes the whole thing tractable.
One programme, not two
The reason they belong together is mechanical. Every major assistant retrieves candidate pages before writing an answer, and that retrieval leans on the same signals as organic ranking — indexability, relevance, authority, internal linking.
You cannot be cited from a page that was never retrieved, and you will not be retrieved if your SEO is weak. AI visibility sits on top of SEO rather than beside it.
Teams that reallocated their whole budget away from SEO to chase AI visibility have generally ended up weaker at both, because they removed the foundation the new work stands on.
The shared seventy percent
Do these once, benefit twice.
- Crawlability and indexation. Pages reachable, allowed, and actually stored.
- Topical relevance and authority. Being a credible answer to the query at all.
- Clear heading structure mapping to real sub-questions.
- Answering the question early. Wins featured snippets and AI extraction with the same edit.
- Specific, verifiable claims. Numbers, dates, named sources — better for readers, snippets and citations alike.
- Internal linking with descriptive anchor text.
- Structured data that matches the visible text.
- Freshness. Maintained pages outperform stale ones in both channels.
- Consolidating duplicates. Helps rankings and produces cleaner citation candidates.
If a proposed piece of work is not on this list, ask which channel it serves before scheduling it.
The AI-only twenty percent
Three additions that do nothing for classic SEO and everything for AI visibility.
Raw-HTML retrievability. AI crawlers do not execute JavaScript. An analysis by Vercel and MERJ covering more than 500 million GPTBot fetches recorded zero JavaScript execution, with the same behaviour from ClaudeBot and PerplexityBot. Googlebot renders, so this is invisible to every conventional SEO audit — and it is frequently the single largest gap on an otherwise healthy site.
Passage-level extractability. Self-contained chunks that survive being copied out of context. Ranking does not require this; citation does.
Third-party corroboration, including unlinked mentions. Classic SEO counts links. Models weigh description — reviews, forum threads, roundups, comparison posts — with no hyperlink required. This is the biggest mindset shift and the slowest to compound.
The SEO-only ten percent
Worth naming so nobody abandons it in enthusiasm for the new channel.
Click-through optimisation. Title tags and meta descriptions exist to convert a ranking into a visit. Assistants do not read them for that purpose, and this remains the cheapest traffic available anywhere in search.
Position mechanics. The specific work of moving from eleven to six — a band where the click-through curve is steep and the return is immediate. AI citation is binary rather than positional, so this has no direct equivalent.
SERP feature capture beyond AI Overviews: image packs, video carousels, local results. Real traffic, no AI analogue.
See both halves in one report
A free scan adds AI visibility to the picture Search Console already gives you — including which competitors get named instead of you.
Run my free scanWhere they genuinely conflict
Three real tensions, each with a resolution.
Comprehensive pages vs single-answer pages. A 4,000-word guide can rank very well and be a poor citation candidate, because no passage in it answers one question cleanly. Splitting it improves citability and can dilute the ranking. Resolution: keep the comprehensive page as the hub, and split out the two or three sub-questions that are being asked as standalone prompts, linking them tightly. You get both.
Client-side rendering. Perfectly acceptable to Googlebot, fatal for AI crawlers. Resolution: there is no trade-off here, only inertia. Server-render the content that matters. Nothing is lost on the SEO side.
Snippet control. The nosnippet and max-snippet directives restrict AI Overview use of your content — and remove ordinary snippets at the same time. Resolution: for the large majority of sites this costs more than it saves. It is a lever that looks like control and behaves like self-harm.
Sequencing one backlog
One list, ordered by leverage across both channels rather than by which channel is fashionable.
- Crawl and index health. Gates everything in both channels.
- Raw-HTML retrievability. Binary, cheap, AI-only, and often the largest single gap.
- Titles and descriptions on high-impression, low-CTR pages. SEO-only, fastest payback in the whole list.
- Consolidate cannibalising pages. Both channels, frequently dramatic.
- Restructure key pages answer-first. Both channels.
- Add specificity and original data. Both channels, and the most durable.
- Corroboration. AI-heavy, slowest, start it early because of that.
Step two typically costs days and step seven costs quarters, which is why step seven should begin in parallel rather than waiting its turn.
Where to start if you have neither
A site with no active SEO and no AI visibility work has one correct first month, and it is not a strategy document.
Verify Search Console and read the Pages report. Load five key pages with JavaScript disabled. Check robots.txt and the CDN for crawler blocks. Run ten buying questions through ChatGPT and Perplexity. That is four tasks, roughly a day of work, and it produces a diagnosis rather than a plan built on assumptions.
In most cases it surfaces one dominant problem — a large unindexed section, a rendering failure, or a category where competitors are named consistently and you are absent. Starting from a named problem is materially different from starting from a channel strategy, and it is the difference between a quarter that moves something and a quarter that produces a roadmap.
Splitting the budget
The question that follows “run one programme” is what proportion goes where. A reasonable default for a team with existing SEO investment:
Sixty percent on shared work. Crawl health, consolidation, answer-first restructuring, specificity, original data. This is where the compounding is and it pays into both channels from a single effort.
Twenty percent on SEO-only work. Titles, descriptions, position mechanics, SERP features. Small allocation, fastest payback, and the reason not to abandon it is that it produces measurable results while the slower work matures.
Twenty percent on AI-only work. Retrievability first — which is cheap and one-time — then corroboration, which absorbs most of this allocation on an ongoing basis.
Two adjustments worth making. If your site is client-side rendered, front-load the AI allocation heavily in the first month, because that single fix is worth more than everything else combined and then the allocation can normalise. And if your organic performance is weak, shift toward the shared and SEO buckets — AI work on a site with no organic presence is optimising a stage you have not reached.
The reporting problem this solves
A quiet benefit of running one programme is that it ends an argument most teams are having. When SEO and AI visibility report separately, every month produces two narratives — one channel up, one flat — and the conversation becomes about which one deserves the next hire rather than about what to fix.
A single report with queries in the top ten, non-branded clicks and share of voice in AI answers describes one system with three symptoms. It also makes the trade-offs visible: if a quarter spent on corroboration moved AI share of voice and left rankings flat, that is a legible outcome rather than a failure in one column and a success in another.
What this looks like against a competitor doing neither
The interesting asymmetry is that most companies are currently doing neither well. Retrievability failures are widespread and largely unnoticed, and consistent third-party description is rare because nobody has historically been rewarded for it.
That means a team executing the shared seventy percent competently, plus the retrievability fix, is ahead of most of its category on both surfaces within a quarter — without doing anything exotic. The window on that will close as the practice becomes standard, which is an argument for starting now rather than for panicking.
How the team changes
Less than the strategic framing suggests, which is worth saying because it lowers the perceived cost of starting.
The content brief gains two fields: the single question this page answers completely, and the one thing it will contain that nobody else has. Fifteen minutes added to a process that already exists.
The weekly review gains one chart: mention rate across a fixed prompt set, by engine, against two named competitors.
Definition of done gains one item: new templates and components render server-side. Once this is established, it stops recurring.
PR gains a target list: the specific third-party domains assistants cite in your category, derived from your own scans rather than guessed.
What does not change: who owns it. One person or team owns organic visibility across both surfaces. Splitting it is how the two backlogs start competing.
Sources
- Vercel and MERJ — analysis of 500M+ GPTBot fetches finding zero JavaScript execution; same behaviour for ClaudeBot and PerplexityBot.
- Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande — “GEO: Generative Engine Optimization”, ACM SIGKDD 2024 (arXiv:2311.09735); ~30–40% relative visibility improvement from statistics, quotations and cited sources.
- Google — 2.5 billion monthly active users for AI Overviews and 1 billion for AI Mode, stated 19 May 2026.
Frequently asked questions
How can I improve both my SEO and AI visibility?
Run one programme rather than two. Around seventy percent of the work is shared — crawlability, indexation, topical relevance, clear structure, specific claims and internal linking. The AI-only additions are raw-HTML retrievability, extractable passages and third-party corroboration. The SEO-only remainder is mostly click-through optimisation and ranking mechanics that have no AI equivalent.
Do SEO and AI optimisation ever conflict?
Rarely, and in three narrow places. Long comprehensive pages can rank well while being poor citation candidates, so splitting them helps AI and can dilute rankings. Client-side rendering is acceptable to Googlebot and fatal for AI crawlers. And snippet-blocking directives protect content from AI Overviews while also removing ordinary snippets, which usually costs more than it saves.
Should I have separate SEO and AI teams?
No. The overlap is around eighty percent and splitting ownership produces two backlogs competing for the same engineering time and two teams making contradictory edits to the same pages. One owner, with occasional engineering support and a route into PR for the corroboration work, is the arrangement that works.
Which should I fix first if I can only do one?
Retrievability, because it is binary, cheap and it gates AI visibility entirely while costing SEO nothing. Load your key pages with JavaScript disabled and confirm the answer is present. After that, work whichever channel your buyers actually use, which for most businesses still means SEO first.
Does improving SEO automatically improve AI visibility?
Partly. Better organic performance improves retrieval, which is the prerequisite for being cited, and it moves Google AI Overviews substantially. It does much less for ChatGPT, which weights third-party corroboration heavily and whose crawler does not render JavaScript. So SEO gains carry over unevenly rather than automatically.
How do I report on both without doubling the work?
Report three numbers per month: queries in the top ten, non-branded organic clicks, and share of voice in AI answers against named competitors. Those cover both channels, cannot be gamed easily, and take the same fifteen minutes to assemble as a single-channel report once the tracking is configured.