Website visibility: 20 tactics to improve it in 2026

How to improve website visibility, in 20 tactics

Website visibility is how findable your content is across every surface where a buyer might encounter it: search results, AI answer engines like Google AI Overviews and ChatGPT, and third-party sites like review platforms and forums. To improve it, work through five layers in order, because each depends on the one below it.

Most sites fail at the first or second layer and spend their budget on the third. A site that isn't indexed does not need a content calendar. A site that ranks but never gets cited does not need more backlinks.

Here are the 20 tactics, grouped by the layer they belong to:

Layer 1, retrievability. Can a crawler reach and read your pages at all?

1. Fix indexation at the reason level, not the page level
2. Serve your content in the HTML, not after JavaScript
3. Decide deliberately which AI crawlers you allow
4. Keep an accurate sitemap and consider an llms.txt
5. Flatten your site architecture

Layer 2, interpretability. Can a machine work out what your pages say and who they say it about?

6. Give every page exactly one job
7. Write self-contained passages under question-form headings
8. Weld entity names to the claims they own
9. Add schema markup that matches what's on the page
10. Make internal links carry meaning

Layer 3, coverage. Do you have pages for the questions buyers actually ask?

11. Map the fan-out, not the keyword list
12. Build the definitional layer
13. Build the comparison and alternatives layer
14. Refresh and prune before you publish anything new

Layer 4, corroboration. Does anything outside your own domain confirm you exist?

15. Get onto the listicles other people write
16. Build presence on review platforms properly
17. Show up where your category argues with itself
18. Publish something only you could publish

Layer 5, conversion. Does the attention turn into pipeline?

19. Match each page to the intent that brought someone to it
20. Track brand search as your leading indicator

Below: what visibility means now and why the old definition broke, a fifteen-minute diagnostic to find your broken layer, all 20 tactics in depth, how to measure any of this when two thirds of searches produce no click, and a 90-day sequence.

What "website visibility" actually means in 2026

Website visibility is how findable your content is across every surface where someone might encounter it, which in 2026 means three distinct places rather than one: traditional search results, AI answer surfaces like Google AI Overviews and ChatGPT, and third-party properties like review sites, directories, and community forums.

That three-way split is new, and it's why most advice on this topic is out of date. For the short definitional version, see what website visibility means. For roughly twenty years, "visibility" and "search rankings" were the same thing, so the tactics were the same thing too. Rank higher, get more clicks, get more traffic. The measurement was clean: position, sessions, conversions.

That chain broke in two places at once.

The first break is that ranking no longer reliably produces a click. According to SparkToro's analysis of Similarweb clickstream data covering US Google searches from January through April 2026, zero-click searches reached 68% of US queries, up from roughly 60% two years earlier. The share of searches generating at least one click fell 9.51 percentage points between 2024 and 2026, a 22.9% relative decline. Pew Research Center's behavioural study of 68,879 real Google searches found that when an AI Overview is present, users clicked a traditional result only 8% of the time, compared with 15% without one.

The second break is that a growing share of the decision happens before anyone visits a website at all. G2's Answer Economy report, based on a March 2026 survey of 1,076 B2B software buyers, found 51% now begin their purchasing process in an AI chatbot rather than a search engine, and that AI chatbots have become the top influence on vendor shortlists. Bain's September 2025 research found 85% of B2B buyers purchase from a vendor list they formed before conducting any search at all.

Put those together and the working definition changes. Visibility is no longer "how high do I rank." It is "how often does my company appear at the moment someone is deciding, on whatever surface they happen to be using, whether or not that produces a click."

A note on the numbers you'll see quoted

Be careful with statistics in this category, including the ones above. The published figures on AI Overview impact vary enormously depending on who measured and how.

Those ranges are not contradictions so much as different methodologies measuring different things: tracked keyword sets versus clickstream panels, all clicks versus clicks to external sites only, different sampling windows. The direction is unanimous across every study. The magnitude is genuinely unsettled.

The practical implication is that you should not build a business case on any single one of these figures, and you should be sceptical of any agency or tool vendor that quotes one number without a methodology. Measure your own site. Your category's numbers will differ from the aggregate, sometimes by a lot.

Diagnose before you fix

Work out which layer is actually broken before running any tactic. Fixing an upper layer while a lower one is broken produces no result and a great deal of wasted budget.

The fifteen-minute diagnostic

Run these six checks before touching anything. Together they will tell you which layer to work on.

1. Is your site indexed at all? Search site:yourdomain.com in Google. If you get nothing, you have a layer one emergency. If you get results, compare the count to your actual page count. A site with 200 pages returning 40 results has an indexation problem, not a content problem.
2. Which specific pages are missing? Open Google Search Console, go to the Pages report, and read the "Not indexed" reasons. "Crawled, currently not indexed" usually means Google found the page and judged it not worth including, which is a quality or duplication signal. "Discovered, currently not indexed" usually means a crawl budget or internal linking problem.
3. Does your content exist without JavaScript? Disable JavaScript in your browser and load your three most important pages. If the copy disappears, some crawlers and most AI retrieval systems are seeing an empty page. This single check catches more visibility problems on modern sites than any other.
4. Are you cited by AI answer engines? Ask ChatGPT, Claude, Perplexity, and Google AI Mode the five questions your buyers actually ask when they're comparing options in your category. Do not ask about your brand by name, that tests recall, not visibility. Record which companies get named and how often. If you never appear, you have a layer two, three, or four problem, and the next checks will tell you which. A GEO audit runs this same check systematically across a larger prompt set.
5. Do you have keyword cannibalization? In Search Console, filter by a keyword you care about and look at which pages receive impressions for it. If three pages split impressions for the same query, you're competing with yourself and Google is picking arbitrarily between your options.
6. Does anything third-party mention you? Search your company name plus your category on Google, excluding your own domain. Count the independent sources. If the only pages describing what you do are pages you wrote, you have a corroboration problem, and it will cap everything else you do.

Layer 1: Make your pages retrievable (tactics 1 to 5)

Everything in this layer is unglamorous and non-optional. It is also where the largest single-day gains in this guide live, because a page that goes from unindexed to indexed goes from zero visibility to whatever it deserves.

1. Fix indexation at the reason level, not the page level

Fix indexation by grouping Search Console's exclusion reasons and treating each reason as its own project, rather than resubmitting individual URLs one at a time. Resubmitting URLs treats symptoms. The reason codes tell you the disease.

Open the Pages report in Google Search Console and sort by the "Why pages aren't indexed" list. Each reason has a different fix:

• "Crawled, currently not indexed." Google fetched the page and chose not to include it. This is a quality judgment. The usual causes are thin content, near-duplicate content across several pages, or a page that adds nothing a hundred other pages don't already say. The fix is consolidation or substantial rewriting, not resubmission. Resubmitting an unchanged page that Google already declined is a request to be declined again.
• "Discovered, currently not indexed." Google knows the URL exists but hasn't crawled it. This is usually a crawl priority problem. The page is buried too many clicks from your homepage, or nothing internal links to it, or your site has enough low-value URLs that Google is rationing attention. The fix is internal linking and reducing crawlable junk.
• "Duplicate without user-selected canonical." Google found several versions of the same content and picked one itself, possibly not the one you wanted. Set the canonical explicitly.
• "Page with redirect" or "Alternate page with proper canonical tag." Usually fine, and usually not worth touching.
• "Excluded by 'noindex' tag." Check every one of these. A noindex directive carried over from a staging environment is one of the most common and most costly launch bugs in web development, and it is completely silent. The site works, looks fine, and is invisible.

One structural note specific to modern site builders. Many platforms generate URLs you never intended to be public: paginated CMS archives, tag pages, filtered views, template preview URLs. These consume crawl budget and dilute your topical signals. Audit what your platform is publishing, and noindex the parts that exist for navigation rather than for readers. On Webflow specifically, the Webflow SEO checklist covers where these URLs usually come from.

2. Serve your content in the HTML, not after JavaScript

Serve the content of every page you want visible in the initial HTML response, because a meaningful share of crawlers and AI retrieval systems either don't execute JavaScript or don't wait for it.

Googlebot does render JavaScript, but rendering is a second, deferred pass with its own queue, which means JS-dependent content is indexed later and less reliably than server-rendered content. More importantly, the crawlers used by AI answer engines are generally less capable than Googlebot. If your value proposition, your product description, and your key claims only appear after a client-side fetch, the systems that increasingly shape buyer shortlists are reading a shell.

The specific things to check:

• Body copy loaded from an API. Common on single-page applications and heavily componentised sites.
• Content inside accordions and tabs. Usually fine if it's in the HTML and hidden with CSS. A problem if it's fetched on click.
• Infinite scroll. Anything below the initial load may never be seen. Provide paginated URLs as well.
• Text rendered inside images. Invisible to everything. Surprisingly common in hero sections and pricing tables.

The test is the one from the diagnostic: disable JavaScript, load the page, and read what's left. If what's left doesn't explain what your company does, you have work to do before any other tactic in this guide will pay off.

3. Decide deliberately which AI crawlers you allow

Decide explicitly whether to allow AI crawlers in your robots.txt, because the default is inconsistent across platforms and the decision has direct consequences for whether you appear in AI answers.

The crawlers worth naming explicitly include GPTBot and OAI-SearchBot from OpenAI, ClaudeBot from Anthropic, PerplexityBot, Google-Extended, and Bingbot. They serve different purposes. Some feed training data, some feed live retrieval for answering questions. Blocking the retrieval crawlers removes you from consideration when someone asks a question your page answers.

There is a real strategic tension here and it deserves an honest treatment rather than a recommendation. Publishers with an advertising model have a genuine argument for blocking: an AI system that reads your content and answers the question without sending a visitor has extracted value and returned none. Companies selling software or services usually have the opposite calculus, because being named in the answer is itself the goal, and the click is a bonus rather than the product.

For most B2B companies, the answer is to allow retrieval crawlers. But make it a decision, check what your robots.txt currently says, and revisit it. Several site platforms have changed their defaults more than once in the past two years without notifying customers.

4. Keep an accurate sitemap and consider an llms.txt

Keep an XML sitemap that lists only canonical, indexable, live URLs, and submit it in Google Search Console. A sitemap does not make Google index a page. It makes discovery faster and more complete, which matters most for new pages and large sites.

The common failure is a stale sitemap. Sitemaps that include redirected URLs, 404s, noindexed pages, or non-canonical duplicates actively harm you, because they signal that your site's own map of itself is unreliable. Regenerate after any migration, any bulk content change, and any URL structure change. Check the sitemap report in Search Console for errors rather than assuming it's fine.

Alongside the sitemap, llms.txt is worth knowing about. It's a proposed plain-text file at your domain root that gives language models a curated, structured summary of your site: what you do, which pages matter, and where the authoritative information lives. Adoption is not universal and it is not yet a ranking factor in any measurable sense. Treat it as cheap insurance rather than a tactic with proven return. It takes an hour, it can't hurt, and if adoption grows you're already there.

5. Flatten your site architecture

Flatten your site architecture so every page you care about is reachable within three clicks of the homepage, because crawl priority and internal link equity both decay with depth.

The practical version of this is not a redesign. It's four checks:

• Check your click depth. Run a crawl with Screaming Frog or your tool of choice and sort by crawl depth. Anything important sitting at depth five or six is being told, by your own architecture, that it doesn't matter.
• Check for orphan pages. Pages with no internal links pointing to them exist only in your sitemap. They get crawled rarely and rank poorly. This is extremely common with older blog posts and landing pages built for campaigns that ended.
• Check your hub pages. Every content cluster should have a hub that links to every page in it, and every page in the cluster should link back. Most sites have the first half and skip the second.
• Check your pagination. Blog archives that go twenty pages deep bury everything after page three. Category hubs and topic pages solve this better than pagination does.

Layer 2: Make your pages interpretable (tactics 6 to 10)

Layer one gets your content read. Layer two decides whether it gets understood, and understanding is what determines whether a passage from your page ends up inside an AI-generated answer. This is the layer most sites skip entirely, because it looks like writing advice rather than technical work.

The mechanic to understand: answer engines do not read your page top to bottom and score it. They break one query into multiple sub-queries, retrieve individual passages that match those sub-queries, and assemble an answer from the passages that most clearly resolve each one. You are not competing as a page. You are competing passage by passage.

That has direct consequences for how copy should be written, and they run against most of what makes for good marketing prose.

6. Give every page exactly one job

Give each page a single primary query to answer, because pages targeting several overlapping queries rank for none of them well and compete with each other for the ones they share.

Keyword cannibalization is the standard name for this and it's usually described as an SEO hygiene issue. It's worse than that in an AI-retrieval world. When two of your pages both partially answer a question, a retrieval system has two mediocre candidate passages instead of one strong one, and the strongest single passage is what wins citations.

Diagnose it in Search Console: filter to a query, look at the Pages report for that query, and see how impressions distribute. Three pages splitting impressions for one term is the signature.

The fixes, in order of preference:

• Merge. If two pages substantially overlap, combine them into the stronger one and 301 the weaker URL. You inherit the links and the history. This is almost always the right answer and almost always the one teams avoid, because deleting a page you paid for feels like waste.
• Differentiate. If both pages have a genuine distinct job, rewrite so the distinction is unmistakable in the title, H1, opening paragraph, and internal anchor text. "What is X" and "How to do X" can coexist. "Guide to X" and "Complete guide to X" cannot.
• Canonicalise. Where duplication is structural and unavoidable, point the canonical at the version you want to win.

A live example of how this bites: a company publishes a definitional glossary entry and a blog post on the same term, both good, both indexed. Neither ranks. After merging into one URL with a 301, the surviving page climbs into the top ten within weeks. The content didn't improve. The signal stopped splitting.

7. Write self-contained passages under question-form headings

Write each section under a heading phrased the way a person would ask the question, and make the section beneath it answer that question completely in 100 to 300 words, without depending on anything earlier on the page.

This is the single highest-leverage writing change for AI visibility, and it costs nothing but discipline.

• Question-form headings. "How much does a Webflow site cost?" rather than "Pricing." "How long does a migration take?" rather than "Timeline." The heading is a retrieval signal. A heading that matches the shape of a real prompt gets matched to that prompt.
• Answer-first construction. Lead each section with the direct answer in the first sentence, then expand. Journalists call this the inverted pyramid. Marketers instinctively do the opposite, building context toward a reveal, which is exactly wrong here. A retrieval system that pulls your first sentence and finds throat-clearing moves on.
• Self-containment. Each chunk should survive being lifted out of the page with zero surrounding context. That means no "as we mentioned above," no "this is why it matters," no pronouns whose antecedent lives three paragraphs up. If a passage only makes sense in position, it can only be read in position, and retrieval doesn't read in position.

The test is mechanical: take any section, paste it into a blank document, and read it. If someone with no other context would understand what it says and who it's about, it will survive retrieval. If not, rewrite it.

8. Weld entity names to the claims they own

State the entity name in the same sentence as the claim about it, every time, rather than relying on pronouns or on context established earlier in the page.

Write "Acme's migration process moves a WordPress site in under six weeks." Do not write "our migration process moves a site in under six weeks," and definitely do not write "the process typically takes under six weeks."

The reason is resolution. When a retrieval system extracts a passage, it needs to know which entity the claim belongs to. If the passage says "we" or "our" or "the company," and the entity name lives in a heading two sections up, the model has a claim floating free of its subject. Some models will resolve it from the page context. Many will drop the passage rather than risk misattribution, and a dropped passage is a lost citation.

This feels repetitive when you write it and reads fine when someone else reads it, because readers skim past the brand name without registering the frequency. Err toward repetition.

The same logic applies to products, features, and proprietary methods. Name them explicitly and consistently. A methodology called "the framework" in half your pages and "our approach" in the other half is two entities that are actually one, and neither accumulates any recognition.

9. Add schema markup that matches what's on the page

Add structured data that describes what the page actually contains, because schema removes ambiguity about page type, authorship, and entity identity in a form machines parse without interpretation.

The types worth implementing for most B2B sites:

• Organization, on the homepage. This is your entity anchor. Include the legal name, logo, URL, and sameAs links to your LinkedIn, Crunchbase, G2, and any other profile that represents you. The sameAs property is how you tell search and AI systems that the entity on those profiles and the entity on your site are the same company. Skipping it means the corroboration you build in layer four may never get connected back to you.
• Article or BlogPosting, on content pages, with a named author, a publish date, and a modified date. Named human authorship is a trust signal that has grown more important as AI-generated content has proliferated.
• FAQPage, on pages with genuine question-and-answer blocks. Google has reduced the rich-result display for this, which caused many teams to drop it. That was a mistake in an AI-retrieval world. The markup still declares "this is a question, and this is its answer" in unambiguous machine-readable form.
• Product and Service, where they apply, including pricing when you publish it.
• BreadcrumbList, which reinforces your architecture.

Two rules. First, the schema must describe what's actually on the page. Marking up FAQs that don't appear in the visible content is a manipulation and gets treated as one. Second, validate it. Google's Rich Results Test and Schema.org's validator both catch the syntax errors that silently disable markup, and silently disabled markup is the normal state of most sites that have never validated.

10. Make internal links carry meaning

Write internal link anchor text that describes the destination page's topic, because anchor text is one of the clearest signals available for what a page is about, and generic anchors waste it entirely.

"Read more" and "click here" tell a machine nothing. "How we handle Webflow migrations" tells it exactly what the destination covers.

Beyond anchor text, three structural habits do most of the work:

• Link from strong pages to weak ones. Your highest-authority pages have equity to distribute. Most sites link outward from their weakest pages and hoard links on their strongest. Find your top pages by traffic and add contextual links to the pages you're trying to lift.
• Build hub-and-spoke clusters properly. One pillar page per topic, linking to every supporting page. Every supporting page linking back to the pillar and sideways to its most closely related siblings. This is how topical authority is expressed structurally rather than just asserted.
• Link in the body, not just the navigation. Navigation links are on every page and are largely discounted as a result. A link inside relevant body copy, with meaningful anchor text, carries far more weight and far more context.

One caution. Do not turn every page into a link farm on the theory that more is better. Links inside prose should be there because a reader might genuinely want that page next. Density beyond that point dilutes each link and reads badly.

Layer 3: Cover the queries buyers actually ask (tactics 11 to 14)

Layers one and two make your existing pages work. Layer three is about whether you have pages at all for the questions that matter. Most sites cover a small fraction of their category's real query surface, because they built pages around keywords with published search volume and ignored everything else.

That approach worked when search was the whole game. It fails now for a specific reason: AI answer engines fan a single prompt out into many sub-queries, most of which have no measurable search volume anywhere. Someone asks a chatbot "what should I look for in a Webflow agency," and the system internally generates a dozen sub-questions about pricing, timelines, partner tiers, migration risk, and post-launch support. Your keyword tool shows zero volume for most of those. They are still the questions being answered.

11. Map the fan-out, not the keyword list

Map the sub-questions a real buyer asks across a purchase decision, then check which ones you have a page for, because coverage gaps are invisible in keyword tools by definition.

A workable method that takes an afternoon:

• Start with the prompts, not the keywords. Write down the ten questions a buyer in your category genuinely asks. Not search queries. Questions. "How do I know if we need this?" "What does it cost?" "How is this different from doing it in-house?" "What goes wrong?" "Who else uses this?"
• Run them through the answer engines. Put each into ChatGPT, Claude, Perplexity, and Google AI Mode. Read the follow-up questions each system suggests. Those suggestions are a direct read on the fan-out, generated by the systems you're trying to appear in.
• Mine the places people ask badly. Reddit, category-specific Slack and Discord communities, review site pros-and-cons sections, and sales call transcripts. People ask questions there in the imprecise language they actually use, which is the language they type into chatbots. Your keyword tool has already normalised that language away.
• Audit coverage. For each question, identify the page on your site that answers it. Not "sort of touches on it." Answers it, in a passage someone could quote. The gaps are your content plan.

Expect the gap list to be uncomfortable. A typical B2B site with fifty blog posts has strong coverage of maybe eight real buyer questions and accidental partial coverage of a dozen more.

12. Build the definitional layer

Publish clear definitional pages for the core terms in your category, because definitional content is disproportionately cited by AI answer engines and it pulls related pages along with it.

The mechanic is worth understanding. When a model answers a question that involves a concept, it needs a source for the concept itself. A page that defines the term cleanly, in a self-contained passage, with the entity welded to the definition, is an efficient thing to cite. And citation is not isolated: once a domain becomes a source a model reaches for on a topic, its adjacent pages surface more readily on related questions.

This is why definitional pages punch above their apparent commercial weight. "What is website migration" is not a bottom-of-funnel term and will never convert directly. It earns citations, and the citations pull your migration service pages into answers where they otherwise wouldn't appear.

How to build the layer:

• Take the terms your category actually uses, including the ones with no measured search volume. Volume is a poor proxy here because these terms get asked conversationally more than they get typed.
• Use a consistent template. Definition in the first sentence. Key takeaways near the top, not the bottom. Question-form subheadings. A short FAQ block. Internal links to the commercial pages the term relates to.
• Put them under a consistent URL path, typically /glossary/, so the pattern is legible to crawlers and to readers.
• Interlink aggressively within the layer. Definitions relate to other definitions. Reflecting that in links builds the topical structure explicitly.

One warning. Definitional pages are the easiest thing to mass-produce with AI and therefore the most commoditised content type on the internet right now. A generic definition adds nothing and will be ignored. What makes one worth citing is specificity: your own numbers, your own examples, distinctions other definitions gloss over, and an honest note about where the term is contested.

13. Build the comparison and alternatives layer

Publish comparison pages, alternatives pages, and category listicles, because these are the formats AI answer engines cite most heavily for commercial questions and the formats buyers explicitly ask for.

When someone asks a chatbot "what are the best tools for X," the model assembles that answer from pages that already list and compare options. If you appear on none of those pages, and you publish none yourself, you are structurally absent from the highest-intent question in your category.

Three page types do the work:

• "Best X" listicles. Include yourself, disclose that you have, and treat competitors fairly. A listicle where the host wins every category is transparently promotional and gets discounted by readers and models alike. Real trade-offs and real caveats are what make the page worth citing.
• "X vs Y" comparisons. These target buyers late in the decision. Be accurate about the competitor, because being caught misrepresenting them costs more than the page earns.
• "Alternatives to X" pages. These capture buyers actively leaving a competitor, which is the highest-intent moment in the entire funnel.

What separates a comparison page that gets cited from one that doesn't is usually specificity on the things everyone else hedges. Publish price ranges when you can. Say who each option is genuinely wrong for. Name the trade-off. A comparison table where every row says "varies" is not a comparison.

14. Refresh and prune before you publish anything new

Update and consolidate what you already have before adding new pages, because refreshing an existing page that already has links and history is typically faster and cheaper than earning a new page into the same position.

Most content programs are structurally biased toward publishing, because new posts are visible work and refreshes are not. This is backwards on the economics.

• Run a content audit. Pull every URL with its organic traffic, keyword positions, backlinks, and last-modified date. Sort into four buckets: performing (leave alone), declining (refresh), never performed but has links (refresh or merge), never performed and no links (prune or merge).
• Refresh with intent, not cosmetics. Changing the date at the top is not a refresh and search engines have been ignoring that trick for years. A real refresh means new information, restructured headings into question form, key takeaways moved up, updated figures, added FAQ coverage, and new internal links to pages published since.
• Prune ruthlessly. Pages that have never earned traffic and have no backlinks are dead weight. They consume crawl budget and dilute topical focus. Merge them into stronger pages with a 301, or remove them. Deleting content feels wrong and is usually right.
• Set a cadence. Quarterly is enough for most sites. Without a cadence this becomes a project someone did once in 2024.

Layer 4: Get corroborated by sources you don't control (tactics 15 to 18)

Everything to this point happens on your own domain. Layer four is about what other people say, and it is the layer with the widest gap between how much it matters and how much attention it gets.

The reason it matters more than it used to is specific to how language models work. A model reading your homepage learns what you claim. A model reading a review site, a Reddit thread, and three independent roundups learns what is true, or at least what is corroborated. When a model assembles a shortlist, the companies that appear across many independent sources are safer to name than the company that appears only on its own website. Bain's finding that 85% of B2B buyers purchase from a vendor list formed before any search is the buyer-side version of the same dynamic.

You cannot write your way onto that list from your own domain.

15. Get onto the listicles other people write

Get your company included in the third-party roundups that already rank for "best X" in your category, because those pages are the raw material AI answers are assembled from.

Find them first. Search "best [your category]", "top [your category] companies", and "[your category] alternatives", and list every roundup on page one and two. Note which ones you appear on and which you don't.

Then work the list:

• Roundups with submission processes. Many have a form or an editorial contact. Use it. This is unglamorous and effective.
• Roundups by agencies or consultancies. These are usually written by someone building their own visibility. A well-researched email explaining specifically why you'd fit, with named clients and a concrete differentiator, converts at a rate that surprises people who've never tried it.
• Roundups on category media. Higher bar, usually requires a relationship or a genuine news hook.
• Roundups that are clearly pay-to-play. Evaluate honestly. Some directories with real traffic are worth the fee. Many are not and exist only to sell listings.

Track this as a proper list with owners and dates, not as an occasional impulse. Ten placements over a quarter changes what models say about your category. One placement changes nothing.

16. Build presence on review platforms properly

Establish and actively maintain profiles on the review platforms your buyers use, because review sites rank exceptionally well for commercial queries and are cited heavily by AI systems evaluating vendors.

For B2B software and services that usually means G2, Capterra, and Clutch, plus category-specific platforms where they exist.

What actually matters:

• Volume and recency beat perfection. A profile with forty reviews averaging 4.4, with recent activity, outperforms one with six perfect reviews from 2023. Recency is a freshness signal and a credibility signal at once.
• Completeness matters more than people expect. Filled-out feature lists, pricing information, screenshots, and category tags are what make a profile usable as a data source. Half-empty profiles get skipped.
• Ask systematically. Build the review request into your delivery process at the moment a client is happiest, usually right after a successful launch or a milestone. Ad hoc asking produces ad hoc results.
• Respond to everything, especially the critical reviews. A thoughtful response to a two-star review is more persuasive to a reader than the review is damaging, and it generates additional text on the page describing how you handle problems.

A note on which platforms to prioritise. The platform your buyers actually consult matters more than the platform with the best domain rating. In B2B software, G2 carries more weight with buyers than most alternatives. In services, Clutch has broad coverage but is less trusted by some segments. Ask your last five customers where they looked. That answer beats any general ranking.

17. Show up where your category argues with itself

Participate in the communities where your buyers discuss your category, because community threads rank well, get cited by AI systems, and are frequently the only unfiltered source available on a topic.

Reddit in particular has become disproportionately visible in search results and is heavily represented in AI training and retrieval. A single thread in the right subreddit can outrank your own service page for a question you spent months writing about.

How to do this without getting removed:

• Participate as a person, not a brand. Use a real account, disclose your affiliation when relevant, and contribute to threads that have nothing to do with your business.
• Answer the question fully before mentioning yourself. The comment that helps someone regardless of what they buy is the one that gets upvoted and cited. The comment that helps only if they hire you gets downvoted.
• Do not astroturf. Fake accounts recommending your company get detected, and the reputational damage substantially exceeds any benefit. Communities are unusually good at spotting this.
• Go where the specific conversations are. Category subreddits, Slack and Discord communities, Indie Hackers, and vertical forums. The broad ones are noise.

This is slow and does not scale, which is exactly why it works. It cannot be automated, so the companies doing it are the ones that decided to.

18. Publish something only you could publish

Publish original research, proprietary data, or a first-hand teardown, because original information is the only content type that cannot be synthesised from existing sources and is therefore the most reliable way to earn both citations and links.

Every generic explainer on your blog competes with a thousand identical explainers and with the model's own ability to write one. Original data competes with nothing.

Realistic formats for a company without a research budget:

• Aggregate data you already have. Anonymised, aggregated benchmarks from your client work or product usage. "We analysed 150 B2B sites and found X" is publishable and nobody else can write it.
• Run a small structured test. Take a question your category argues about, test it properly, and publish the method alongside the result. Sample sizes can be modest if the method is transparent.
• Survey your own audience. Even a few hundred responses produces citable figures if you publish the methodology honestly.
• Do a teardown. Analyse twenty real examples of something in your category and extract the pattern. Labour-intensive, not intellectually difficult, and genuinely useful.

The two rules that decide whether it works: publish the methodology, including sample size and limitations, and put the headline finding in a self-contained sentence near the top so it can be extracted and cited without the reader needing the rest of the page.

Layer 5: Convert the visibility you earn (tactics 19 and 20)

Visibility that produces no pipeline is a reporting line item. This layer is short because it is not the subject of this guide, but skipping it is how companies end up with a traffic chart going up and a revenue chart going sideways.

19. Match each page to the intent that brought someone to it

Give every page a call to action proportionate to the intent behind the query it ranks for, because mismatched CTAs waste the visibility you spent months earning.

Someone arriving on a definitional glossary page is at the beginning of a problem, not the end of a decision. A "book a demo" button is the wrong ask, and the bounce it produces teaches you the page doesn't work when actually the page is fine and the ask is wrong. Offer the next piece of information instead: the related guide, the deeper explainer, the tool.

Someone arriving on a comparison or alternatives page is at the end. Ask for the meeting.

The practical version is to sort your pages by intent, assign one primary CTA per intent tier, and stop putting the same demo button on all of them.

20. Track brand search as your leading indicator

Track branded search volume and direct traffic over time, because in a zero-click environment they are the clearest available evidence that AI-surface visibility is working.

Here is the measurement problem. When someone reads about you inside a ChatGPT answer and then searches your company name and visits your site, your analytics records that session as direct or branded organic. The AI conversation that caused it is invisible. Almost all attribution models will credit the wrong thing.

Branded search volume is the workaround. If your branded query volume and direct traffic rise while your non-branded organic is flat, something is putting your name in front of people somewhere you can't see. In 2026 that somewhere is usually an answer engine.

Watch the ratio rather than the absolute number, and watch it monthly rather than weekly. Noise dominates at shorter intervals.

How to measure website visibility when clicks are unreliable

Measure website visibility across four separate reports rather than one, because no single metric survives the shift to zero-click search and AI answers.

The old dashboard was sessions, rankings, and conversions. Sessions now understate reach, because a growing share of impressions never produce a click. Rankings now overstate value, because position one with an AI Overview above it performs unlike position one without. Neither is wrong. Both are incomplete.

1. Impressions and position, from Google Search Console. Impressions are now the more honest reach metric, because they count the times you appeared regardless of whether the click happened. Track impressions and average position together. Impressions rising while clicks fall is not failure. It is the market condition, and it is measurable.

Use the comparison view rather than the absolute view. Your year-over-year click decline means little without knowing whether your category's declined more or less.

2. AI citation tracking. Tools including Profound, Scrunch AI, Peec, and Otterly track how often your brand appears in AI-generated answers for a set of prompts you define. The methodologies vary and the category is young, so treat absolute numbers as directional and watch the trend instead.

The free version of this is a spreadsheet. Define twenty prompts your buyers would actually type. Run them monthly across ChatGPT, Claude, Perplexity, and Google AI Mode. Record whether you were named, in what position, and who else was. This takes an hour a month and gives you a real time series. Most companies do neither the paid nor the free version and simply guess.

3. Branded search and direct traffic. Covered in tactic 20 above. This is your proxy for the influence you cannot see directly.

4. Referral traffic from AI platforms. Traffic from chatgpt.com, perplexity.ai, claude.ai, and similar sources appears in GA4 under referrals. The volume is usually small and the intent is usually high, because someone who clicks through from an AI answer has already been told you're relevant. Segment it and look at conversion rate rather than volume. A hundred sessions converting at eight percent matters more than five thousand converting at nought point four.

A realistic timeline

Set expectations against these ranges, which reflect typical B2B sites rather than best cases:

The AI citation row carries the widest error bars, because retrieval indexes update on schedules the platforms don't publish and don't guarantee. Some changes surface within days. Some take a quarter.

A 90-day sequence

Run the work in dependency order rather than by perceived importance, because upper-layer work performed on a broken lower layer produces nothing.

Days 1 to 14: diagnose and fix retrievability. Run the six-check diagnostic. Fix indexation by reason code. Check the JavaScript rendering of your top pages. Audit robots.txt for AI crawler directives and make that a deliberate decision. Regenerate and resubmit the sitemap. Do not write anything new during this window.

Days 15 to 30: fix interpretability on the pages you already have. Take your ten highest-impression pages and rework them: question-form headings, answer-first openings, self-contained sections, entity names welded to claims, key takeaways moved to the top, and an FAQ block where it fits. Add and validate schema. Resolve any cannibalization you found.

This window usually produces the largest measurable gain in the whole ninety days, because it improves pages that already have authority rather than starting new ones from zero.

Days 31 to 60: close the biggest coverage gaps. Run the fan-out mapping exercise. Pick the five gaps with the clearest commercial relevance and build those pages. Start the definitional layer for your core terms. Begin the refresh queue on declining pages.

Days 61 to 90: build corroboration and start measuring properly. Compile the third-party roundup list and start outreach. Complete your review platform profiles and build the review request into delivery. Publish one thing only you could publish. Set up the AI citation tracking spreadsheet or tool. Establish the branded search baseline.

Day 91 onward. This is a cadence, not a project. Quarterly content audits, monthly citation tracking, continuous placement outreach.

Six mistakes that waste the most budget

1. Publishing before fixing. New content on an unindexed or unrenderable site is money set on fire. The order in the ninety-day plan is not arbitrary.
2. Optimising for volume that no longer converts. A term with 5,000 monthly searches that now resolves entirely in an AI Overview is worth less than a term with 200 searches that still produces clicks. Check the SERP for a term before committing to it, rather than reading the volume column and stopping there.
3. Treating AI visibility as a separate project. Answer engine optimization is not a bolt-on channel. It is a set of structural and writing decisions made during the build. Retrofitting it costs several times what building it in costs.
4. Measuring only what your analytics can see. If your dashboard shows sessions and conversions and nothing else, you are blind to the surface where a growing share of the decision now happens. That does not mean the influence isn't there.
5. Producing volume instead of substance. With generative tools, anyone can publish forty posts a month. Industry research compiled for 2026 found 43% of B2B marketers now struggle to differentiate their content in a market saturated with mass-produced output. Volume is no longer scarce. Original information is.
6. Confusing visibility with demand. If nobody wants what you sell, being highly visible surfaces that faster. Visibility amplifies whatever is already true about your offer, in both directions.