--- title: "The Influenceable Web" description: "Most of what decides whether AI recommends you happens on surfaces you don't own." url: https://research.momenticmarketing.com/state-of-the-influenceable-web-2026 author: "Tyler Einberger" publisher: Momentic Research published: "May 2026" --- # The Influenceable Web *Most of what decides whether AI recommends you happens on surfaces you don't own.* ## The Influenceable Web *Momentic Research · May 2026* Most of what decides whether AI recommends you happens on surfaces you don't own. How that part of the internet works, and how to earn a place on it. ## The web you can measure is shrinking. *What it is* The Influenceable Web is the part of the internet where you can still see who showed up, what they did, and whether it helped the business. For years that meant your own website. Now the decisions happen in places you can shape but never fully own, and the clean reporting you used to have is fading. ## The Influenceable Web in four parts. *What it covers* - What it covers. Communities, reviews, editorial, earned mentions. The outside surfaces AI reads to decide who to trust. - Why it's growing. AI answers pull from third-party sources far more than from any single brand website. - Why it's hard. You can influence these surfaces, but you can't own them or measure them cleanly. - What to do. Make work worth citing, get it onto those surfaces, judge it by presence over time. ## The measurable web contracted last year. *What the clickstream shows* Hand-pulled Similarweb clickstream across ~100 deliberately chosen, browser-based websites, tracked monthly since March 2022. The audiences moved into in-app surfaces, the ones older analytics decks called “dark.” **3%** — contraction in the Influenceable Web, 2025 vs 2024 - websites hand-tracked: ~100 - monthly since: 2022 *Source: Momentic Research · Similarweb clickstream, Apr 2022 – Apr 2025.* ## The buyer's path went dark. *The shift* A path used to run through surfaces you could track: search, your website, conversion. Today the same buyer might start in iMessage, move through the Reddit app, validate the answer in ChatGPT, and transact in a marketplace app. None of it shows up in a standard analytics stack. ## The same purchase, two eras apart. *Then vs now · pick an era* **A buyer wants to choose** Same intent in both eras: someone needs to pick a product and wants to get it right. What changed is the route they take to an answer, and how much of it you can see. **Then** - *They searched* — The journey opened on a search engine you monitored. Query data, impressions, the whole funnel entrance was visible. - *They landed on your website* — You saw the visit, the path through your pages, the source. You owned the stage and the analytics that described it. - *They converted* — The result tied back to the visit. One buyer, one trackable line from first click to purchase. **Now** - *They ask an assistant* — The journey opens inside ChatGPT, Perplexity, or an AI Overview. You never see the question. - *It pulls from surfaces you didn't write* — The recommendation is assembled from reviews, threads, and articles, most of them outside your domain. - *They transact in an app* — Checkout happens in a marketplace or in-app surface that reports nothing back to you. The line from intent to sale is gone. *Source: Momentic Research · The Influenceable Web.* ## A model knows you two ways, and the two work on different timescales. *How AI knows about you* - Channel 1 · Training: pages the model absorbed when it was built. Locked in until the model retrains. Earned slowly, over years. - Channel 2 · Live search: pages the model pulls through an index at question time. Fresh work and distribution pay off here in weeks. This is the channel you can move. - The search layer is fragmenting. Perplexity and OpenAI are building their own indexes, so the surfaces you earn need to work for more than Google and Bing. ## Most of what AI cites about you, you don't own. *Where the answer forms* When researchers audited the sources AI named for unbranded questions about a major brand, the brand's own website was about 2% of the picture. By one framework, an SEO team controls only two of the seven functions that drive answer-engine outcomes. The rest live in PR, community, product, and the open web. **90%** — of AI citations about a brand come from off-site - the brand's own website: ~2 % - of 7 levers SEO owns: 2 *Source: VaynerMedia brand audit; Eli Schwartz's seven-function framework. Directional, single-brand.* ## Your own website is the smallest slice. *Share of AI citations by source* Across unbranded queries about one brand, independent articles and community surfaces carried the answer. The page you control most was a rounding error. | | | |---|---:| | Independent articles | 54% | | Social & community | 25% | | Other sources | 19% | | Your own website | 2% | *Source: VaynerMedia brand audit, unbranded queries. Directional, single-brand.* ## A handful of surfaces decide the answer. *Where it lives · tap a surface* *Source: Share of AI citations by source type, unbranded queries. VaynerMedia brand audit; directional.* ## Getting cited is mostly a distribution problem. *How you get there* More posts on your own website won't get you cited. Placement will: take work good enough to cite, put it in front of the surfaces that already get cited, and do it again. One asset becomes many placements, so most of the effort goes into the spread, not the draft. ## One asset, many placements. *Press distribute* *Source: Around 80% of the channel is placement rather than writing. 94% of “best [software]” queries had a community thread as the single most influential link.* ## Distribution needs something worth carrying. *What you distribute* A distribution engine moves whatever you give it. The brands that earn citations publish things a competitor can't reproduce: original data, real testing, a point of view worth quoting. Templated posts and AI-spun filler give the engine nothing to move, and pull your brand into the same blur as everyone writing the same thing. ## Three things models keep citing. *Cargo worth moving* - First-party data: numbers only you have, from your own customers, tests, or research. - Real testing: you used the thing, took it apart, measured it. A model can't fabricate that. - A clear point of view: an argument worth quoting rather than a summary of what everyone already says. ## The foundation and the wider web do different jobs. *Your own website still matters* - Your website does this: proves you're real and credible, gives the model clean specs to quote, anchors your entities and schema. - The wider web does this: carries most of the citations, forms the recommendation, vouches for you in third-party words. ## You won't tie a citation to a sale. Measure it anyway. *How to measure it* The clean attribution is gone. You can't trace a single AI mention to a dollar, and any tool that promises otherwise is selling you a story. What you can do is track whether you show up, how often, and which way it's trending, then prove value the way brand work has been reported for decades. ## Presence, accuracy, lift. *Three things you can track* - Track presence. Are you named in the answers that matter, and is that share rising or falling over time? - Read the why. When AI describes you, is it accurate and on-message, or working from something stale? - Prove value by lift. Holdouts, correlation, and brand-search trends rather than a fabricated cost-per-citation. ## Three rules for the dataset. *How I read the clickstream* - Blend the sources. Clickstream, platform data, server logs. No single source is reliable enough on its own. - Track relative movement. Month over month is where the signal lives rather than the absolute number. - Read the category shifts. If Video & Streaming climbs in your audience while Reference & Forums stays flat, that's the trend worth acting on. ## Five moves to earn the Influenceable Web. *What to do Monday* - Map your surfaces. Find the specific communities, reviews, and publications AI cites in your category. Start there. - Make something worth carrying. Original data or testing rather than another summary. - Show up honestly. Earn a presence in those communities and reviews. You can't buy or fake your way in. - Distribute, then distribute again. One asset, many placements. Spend most of the effort here rather than on the draft. - Track presence rather than a vanity ROI. Watch whether you appear and how it trends. > Months of earned off-site presence build a position competitors can't buy back. > Earning a position on the surfaces AI cites takes months, sometimes years. Which is why I run this dataset monthly instead of waiting for an annual report. The shifts that decide who shows up next year are in the data now. > — Tyler Einberger, Momentic ## Most of what AI ends up citing forms off your own website now. *Where this leaves you* Keep your own pages clean and quotable, then put your effort into work worth citing and getting it onto the surfaces that shape the answer. That part of the web is measurable, and it's where the recommendation gets made.