How to improve brand visibility in AI search: what moves the needle

What brand visibility in AI search means

Brand visibility in AI search is how often, and how accurately, AI assistants such as ChatGPT, Perplexity, Gemini, Copilot and Google’s AI Overviews mention, recommend or cite your brand when buyers ask questions in your category. It is a share of relevant answers, not a position, and it depends heavily on sources you do not own.

Four different things get called visibility, and they respond to different work. Separating them is the first step, because a team that only counts citations of its own URLs will miss most of what assistants say about it.

Kind	What it is	What mostly moves it
Mention	Your brand is named in the answer	What third-party sources say about you
Recommendation	Your brand is suggested as an option for the buyer’s need	Mentions on comparison lists, reviews and forums, with the right positioning
Citation	One of your URLs is linked as a source	Ranking for the sub-queries, and passages that stand alone
Accuracy	What the answer says about you is correct	Consistent facts on your site and on third-party profiles

Framework for this guide

Citations are the part most guides cover, and the on-page work for them is set out in full in the guide to AI search optimization and what gets retrieved. This guide is about the other three, which is where most brands are weakest. A buyer asking “what are the best expense tools for a 50-person company” often receives a list of brand names drawn from comparison articles, review sites and forums, as the source studies below show. Your own blog may never be cited in that answer, and you can still be the first name on the list.

What drives brand visibility in AI search: the evidence

The strongest published predictor of brand visibility in AI search is how often the brand is mentioned across the web, followed by ranking well in Google. Backlinks and Domain Rating, the traditional SEO authority signals, correlate only weakly. These are correlations, not proof of cause, but they are the best evidence available.

Ahrefs studied 75,000 brands across ChatGPT, Google AI Mode and AI Overviews using Spearman correlation. For AI Mode, branded web mentions correlated at 0.709 with AI brand visibility, branded anchors at 0.628, branded search volume at 0.466, and Domain Rating at 0.285. YouTube mentions correlated most strongly of any factor tested, at about 0.737 across the three platforms. Link metrics such as backlink counts showed only very weak correlations, and Ahrefs notes that none of this means improving a metric will automatically raise AI visibility.

Branded web mentions 0.709

Branded anchors 0.628

Branded search volume 0.466

Domain Rating 0.285

Ahrefs, 75,000 brands, December 2025

A separate study by Seer Interactive took 10,000 questions drawn from finance and SaaS keywords, generated answers with OpenAI’s GPT-4o, and compared brand mentions with search rankings. Google page-one rankings correlated with mentions at about 0.65, which the authors called “the strongest predictor of brand mentions in AI answers”. Bing rankings correlated at about 0.5 to 0.6, and backlinks showed a weak or neutral effect.

What transfers from SEO, and what does not

Read together, the two studies sort SEO into three groups. What transfers: ranking well for the questions buyers ask, because assistants that search start from the same results. What transfers weakly: link building for its own sake, since backlink counts barely predicted visibility in either study. What does not exist in classic SEO at all: being named, in your own words, on the pages that assistants read when they answer a category question. That third group is the gap most programmes have.

The caution is that correlation runs both ways. Well-known brands get mentioned more and also get recommended more, so some of the relationship is fame causing both. The practical reading is narrower and safer: being absent from the sources assistants read makes you absent from the answers, and that absence is something you can fix.

How to set a baseline for brand visibility in AI search

A baseline for brand visibility in AI search is a fixed set of buyer questions, run in several assistants in clean sessions, with each answer recorded for whether you are mentioned, recommended or cited, who else is named, and which domains are cited. Building it is a one-off job. The recurring cost is the rerun, and it grows with every question, run and assistant you add: 40 questions run three times in three assistants is 360 answers to record each month.

Write questions the way a buyer asks them, without your brand name, across the stages of a purchase. Thirty to fifty questions is enough for a first baseline.

Stage	Question pattern	Example for an expense tool
Problem	How do I solve [problem] for [segment]?	How do I stop staff losing receipts at a 50-person company?
Category	What are the best [category] for [segment]?	What are the best expense management tools for small businesses?
Comparison	[Competitor A] vs [Competitor B]?	Expensify vs Ramp for a UK company?
Alternatives	Alternatives to [competitor]?	What are the alternatives to Expensify?
Brand facts	Does [brand] do [capability]? How much is [brand]?	Does [your brand] integrate with Xero?

Template; replace the examples with your own category

Run each question at least three times per assistant in a logged-out or temporary session. Answers vary a great deal between runs: in the AirOps study of ChatGPT, only 2.3% of page and query combinations were cited in all three runs. Record five fields per answer: mentioned (yes or no), recommended (yes or no), your position in any list, competitors named, and every cited domain.

The full tracking method, including tools, cadence and how to read noisy results, is covered in how to track brand mentions in AI search. For this guide the baseline has one more job: it is the raw material for the source-gap analysis below.

How to run a source-gap analysis

A source-gap analysis lists the domains that AI assistants cite in answers where your competitors appear and you do not, then groups those domains by type so each type gets a specific action. It turns a vague goal, “be more visible”, into a list of named pages to get onto.

1. From the baseline, filter to answers where at least one competitor is mentioned and you are not.
2. List every cited domain in those answers, and count how many answers each domain appears in.
3. Classify each domain: review platform, comparison or “best X” article, forum or Reddit thread, video, news or trade publication, documentation or marketplace listing, encyclopedia.
4. For each of the top 20 domains, open the cited page and note whether you are on it, what it says about you if so, and who controls it.
5. Assign an action per type, using the table in the next section.
6. Rerun the same questions after the work and count whether those domains now mention you.

A worked example

The source-gap checklist

The checklist below covers one full pass of the analysis, from a clean baseline to a recount. Ticks are kept in your browser, and the list copies as Markdown for a project ticket.

Baseline

30 to 50 buyer questions written without the brand name, across problem, category, comparison, alternatives and brand-fact stages Each question run at least three times per assistant in logged-out or temporary sessions Five fields recorded per answer: mentioned, recommended, list position, competitors named, cited domains

Counting the gap

Answers filtered to those where a competitor is mentioned and you are not Every cited domain in those answers listed and counted Each domain classified: review platform, “best X” article, forum, video, publication, marketplace, encyclopedia Top 20 cited pages opened, with your presence and what they say about you noted

Acting on it

One owner and one action assigned per source type List authors briefed with accurate, dated product facts Review requests sent to real customers on the platform cited most Wrong facts corrected openly, from a named company account, with a source

Rechecking

Same questions rerun after the work, in the same conditions Mention rate, share of voice and source coverage compared with the baseline

Ticks are saved in your browser.

How to get onto the sources AI assistants already cite

Getting onto the sources AI assistants already cite means earning accurate mentions on the review platforms, comparison articles, forums, videos and reference sites that appear in your category’s answers. Each source type has its own rules, and the wrong approach on several of them does more harm than absence.

Source type	What the evidence shows	Action
“Best X” lists and comparisons	43.8% of page types cited in ChatGPT’s sources across 750 prompts (Ahrefs)	Brief the authors of the lists that appear in your answers; supply accurate facts
Review platforms	Check your own panel: some categories lean on them heavily	Ask real customers for reviews on the platforms your answers cite
Reddit and forums	Often retrieved, cited far less; share swings sharply	Answer questions openly as the company; never astroturf
YouTube	Strongest correlate with AI brand visibility (Ahrefs)	Demos and comparisons on the questions your panel asks
Wikipedia	59.2% citation rate in ChatGPT, against 2.4% for YouTube pages (AirOps)	Do not edit your own article; propose changes on its talk page

Ahrefs, AirOps, Semrush and Wikipedia, linked in the text

Comparison and “best X” articles

Ahrefs analysed 750 top-of-funnel prompts across software, products and agencies and found that “best X” blog lists made up 43.8% of all page types among ChatGPT’s source links. Higher placement on third-party lists correlated with more ChatGPT recommendations, and 35% of the cited lists sat on low-authority domains, so these are not only the big publishers. The practical move is a short, factual briefing to each list author that appears in your source-gap analysis: what the product does, who it is for, current pricing, and what it does not do. Paying for placement without disclosure is a trust risk that outlasts any gain.

Reddit and forums

Reddit’s role is large and unstable. Semrush tracked more than 230,000 prompts over 13 weeks in 2025 and found Reddit cited in close to 60% of ChatGPT responses in early August, falling to around 10% by mid-September. Ahrefs found that ChatGPT retrieves Reddit pages often but cites them rarely: Reddit results were cited 1.93% of the time they were retrieved, against 88.46% for general search results. Treat forums as a place where answers are formed and corrected, not as a citation channel. Answer questions in your category from a clearly identified company account, and correct wrong facts about your product politely and with a source.

Wikipedia

Wikipedia matters where your company qualifies for an article: in the AirOps study, Wikipedia pages had a 59.2% citation rate in ChatGPT, against 2.4% for YouTube pages. YouTube’s strong correlation with brand visibility and its low citation rate are not a contradiction: a video can shape what is said about a brand without being the page that gets linked. Wikipedia’s rules are strict. The conflict of interest guideline says editors with a conflict of interest are “strongly discouraged from editing affected articles directly” and may instead propose changes on talk pages, and that paid editors must disclose who is paying them. If your article is wrong, propose a sourced correction on the talk page. If you do not have one, the route is coverage in independent publications, not writing it yourself.

How to make your brand facts consistent everywhere

Brand facts are consistent when your name, category, description, pricing model, key integrations, target customer and limitations read the same on your site and on every third-party profile. Assistants assemble answers from many sources, so a contradiction between two of them becomes a hedge or an error in the answer.

Start with a single brand fact sheet that the whole company uses as the source of truth, then make every profile match it.

Then audit each profile against the sheet: LinkedIn, Crunchbase, G2, Capterra, app marketplace listings, partner directories, your documentation and your own About and pricing pages. Mark each field as matching, outdated or missing, and fix the outdated ones first, because an outdated fact repeated in several places is the version most likely to reach an assistant from more than one source.

On your own site, Organization structured data with sameAs links to those official profiles states which profiles belong to you. It does not by itself earn citations, but it removes one source of ambiguity; the Schema Validator checks that the markup on a page does not contradict itself.

When an assistant gets your brand wrong

1. Record the exact wrong statement, the assistant, the date and whether it cited a source.
2. If it cited a source, find the claim there. Correct it at the source: update your page, ask the publisher, or fix your profile.
3. If it cited nothing, search the web for the wrong claim in quotation marks to find where it is repeated, and correct the copies you can reach.
4. Publish the correct fact plainly on your own site, in one sentence a model can quote.
5. Recheck monthly. Searched answers change once the sources change; answers drawn from a model’s training change only when a new model is trained.

The difference between those two kinds of answer is sharpest in ChatGPT, which answers either from a live search or from training, and only one of those can be corrected quickly.

How to make your own site the source worth citing

Your own site becomes the source worth citing when it holds facts other pages do not: original data, precise product facts, honest comparisons and specific answers to buyer questions. Assistants choose between passages, so the page with the most specific answer to a sub-question has an advantage regardless of brand size.

Not every page is worth this effort. Assistants are asked questions, so a page that answers one is a candidate; a page that exists to convert someone already convinced is not.

Page type	Worth the effort	Why
Original research or data	Yes, highest value per page	A number nobody else has gives an assistant a reason to cite you
Comparison and alternatives pages	Yes	Heavily asked, and specific enough to be citable if honest
Explainers and guides	Yes	They answer questions in the form assistants receive them
Pricing and product facts	Yes, briefly	Brand-fact questions need a one-line answer you control
Case studies	Rarely	Specific to one client; the question that would surface them is rare

Editorial judgement for this guide

Write those pages so each section stands alone and opens with its answer. Microsoft’s guidance on inclusion in AI search answers notes that assistants “can often lift these pairs word for word” when a page pairs a direct question with a clear answer, and warns against hiding key details in tabs, PDFs or images. The AI Citability Checker tests whether a passage reads on its own.

Comparison pages deserve particular care. A comparison that admits where a competitor is the better choice is more useful to a buyer and more likely to be read as a fair source. One that claims you win on every row reads as an advertisement and gives any summary nothing balanced to quote.

How to use original data to earn brand mentions

Original data earns brand mentions because an assistant answering a question about a number has few sources to choose from, and the source of the number is usually named. A small study with a clear method, published in plain HTML, can be cited more reliably than a guide competing with a hundred similar guides.

The academic evidence points the same way. The GEO paper presented at KDD 2024 tested rewrites on a benchmark of 10,000 queries and found that adding statistics, adding quotations and citing sources were among the most effective changes, with gains of up to 40% on its visibility measure. That was a research setting, not a live engine, but the direction is consistent with what large citation studies show.

1. Pick a question your buyers ask that has no good number. Check your panel for questions where assistants hedge or cite old data.
2. Choose a method you can defend. A customer survey, anonymised product usage data, or a structured test of tools in your category.
3. Publish the method beside the result: sample size, dates, how respondents were chosen, and the exact question wording.
4. Put each headline finding in one sentence with the number, the population and the date, so it can be quoted alone.
5. Brief the publications and list authors from your source-gap analysis with the finding. Third-party coverage is what turns your number into a mention.

A worked example: one quotable finding

Should you publish your own “best X” list?

Publishing your own “best X” list can help brand visibility in AI search, because such lists are heavily represented in assistants’ sources, but it carries a trust cost that grows the more one-sided the list is. Make one only if you can write it fairly.

In the Ahrefs study of best lists, 34% of software brands’ first-party mentions in ChatGPT came from their own blog lists. The author, Glen Allsopp, concluded that SaaS companies and agencies should publish such lists, since the data showed no apparent ranking or citation downside, while warning that many are overly self-promotional, fail to link to alternatives, and may suffer from poor user signals later. He said Ahrefs would keep comparison posts to less than half a percent of its blog content and link directly to alternatives.

Can you describe each competitor accurately, including where it is the better choice?

No

Do not publish the list. If the only honest version puts you first on every criterion, it will read as an advertisement. Publish a comparison page with stated criteria instead.

Yes

Do you already have more than a handful of these pages?

Yes

Improve the lists you have rather than adding more. A growing set of self-ranked lists starts to look like a content farm in your own category, which damages the credibility of the rest of the site.

No

Publish it. State the criteria and the date, link to the alternatives, and update the list when products change.

Mistakes that hold brand visibility back

The mistakes that most often hold brand visibility in AI search back are measuring only your own citations, chasing links rather than mentions, inconsistent brand facts, and blocking the crawlers that feed search answers. Each one is easy to spot once you look.

• Counting citations of your URLs as the whole picture. Most category answers name brands without citing their sites. Track mentions and recommendations as well.
• Testing in a logged-in account. Memory and past conversations can bias the answer toward a flattering result. Use clean sessions.
• Asking leading questions. A question that includes your brand name guarantees a mention and tells you nothing.
• Positioning copy instead of facts. A coined category name nobody else uses gives assistants nothing to match, and they describe you with someone else’s words instead.
• Blocking search crawlers by accident. A blanket rule against AI bots often blocks OAI-SearchBot or PerplexityBot along with the training crawlers.
• Astroturfing forums or buying undisclosed reviews. It is the opposite of an accurate description, and if it is exposed, the exposure becomes one more page assistants read about you. Google’s AI optimization guide makes the same point from its side: seeking inauthentic mentions across the web “isn’t as helpful as it might seem”.
• Reacting to one run. Answers vary between runs and model versions. Read movement over two or three monthly panels.

How to tell whether brand visibility in AI search is improving

Brand visibility in AI search is improving when your mention rate and share of voice on a fixed question panel rise over two or three months, when the description of your brand gets more accurate, and when branded search and assistant referrals rise alongside. No single number is enough.

• Mention rate: answers mentioning you divided by total answers in the panel.
• Share of voice: your mentions divided by the mentions of you and a fixed set of competitors, worked through in the calculator below.
• Accuracy: of the brand-fact questions, how many answers are fully correct.
• Source coverage: of the top 20 domains from the source-gap analysis, how many now mention you accurately.
• Branded search volume in Google Search Console, a slower but independent signal that more people have heard of you.
• Citation reports, such as the AI Performance report in Bing Webmaster Tools, which shows citations and grounding queries across Microsoft Copilot and Bing’s AI summaries.

How to calculate share of voice and the gap to a target

Share of voice and mention rate answer different questions, so calculate both. Peec AI’s metric documentation defines share of voice as your brand mentions divided by the total brand mentions in the answers, times 100, and treats visibility, the share of answers that mention you, as a separate measure. The formulas in the calculator below are: mention rate = answers mentioning you / total answers x 100; share of voice = your mentions / (your mentions + tracked competitor mentions) x 100; and, holding competitor mentions constant, the mentions needed for a target share t are t x competitor mentions / (100 – t).

Enter the counts from one run of your question panel.

Answers in the panelAnswers that mention youMentions of tracked competitors (all of them)Answers that mention the top competitorTarget share of voice%

Your mention rate

15.0%

Top competitor’s mention rate

45.0%

Your share of voice

12.0%

Mentions needed for the target share

33

Holding competitor mentions constant. Above the number of answers, the target is out of reach

Additional answers that must mention you

15

Defaults are the hypothetical expense-tool panel from the source-gap example. Count one mention per brand per answer. Change any number.

With the defaults, the mention rate is 18 / 120 = 15.0%, the top competitor’s is 54 / 120 = 45.0%, and share of voice is 18 / (18 + 132) = 12.0%. Reaching a 20% share with competitors unchanged needs 20 x 132 / 80 = 33 mentions, 15 more answers than today. The two measures diverge in a useful way: if competitors start appearing in more answers, your share of voice falls even when your mention rate holds, which tells you the category answers are getting more crowded rather than that you are losing ground.

Share of voice is most useful split up. Similarweb’s guide to AI share of voice suggests reading it as a trend over time, against your top competitors and across topics. For a B2B panel, the natural topic split is the buying stage from the question template: a brand can hold a strong share on brand-fact questions and a weak one on category questions, and only the split shows which stage the source-gap work should target.

Assistant referrals in analytics are worth watching but undercount, since many answers are read without a click. The broader problem of reporting a channel whose influence is mostly invisible is covered in measuring organic when attribution breaks.

A 90-day plan to improve brand visibility in AI search

A 90-day plan to improve brand visibility in AI search runs in three phases: a baseline and fact audit in the first week, source-gap work and page fixes in the first month, and original data plus a second measurement by the end of the quarter.

Week 1

Baseline and facts

Write a 30 to 50 question panel and run it three times per assistant. Write the brand fact sheet. Check robots.txt for blocked search crawlers.

Month 1

Close the biggest gaps

Run the source-gap analysis. Fix outdated facts on every profile you control. Brief the authors of the top comparison lists. Start a review request to real customers on the platform cited most. Fix the ten most-asked questions on your own site.

Quarter 1

Create a reason to be named

Publish one original data study with its method. Brief the publications in your source list with it. Rerun the panel at month two and month three and compare mention rate, share of voice and accuracy with week 1.

Keep it inside the work you already do. Most of the inputs are shared with SEO, content and PR, and a separate AI visibility initiative tends to produce a document rather than a change. The content side of this, getting a study in front of the people who will cite it, is the same distribution problem described in B2B content marketing as a distribution problem. The vocabulary that surrounds all this, from GEO to LLM SEO, is untangled in what LLM SEO means and what is actually different.

Where a team wants the panel, the source-gap analysis and the fixes run as one piece of work, that is the scope of the AI search visibility service. Before any of it, the Entity Extractor is a quick way to see whether your own key pages state the facts you want repeated, or only gesture at them.