How we‘re driving AI visibility at Semrush

Over the previous 12 months, we have gone from invisible in AI solutions for our class to persistently displaying up for the prompts consumers truly use. That is the playbook — up to date with what practically a 12 months of working it has taught us.

It began with a brutal wakeup name.

Simply weeks after launching Enterprise AIO and the AI Visibility Toolkit, I requested ChatGPT a easy query about AI monitoring instruments. It named each competitor — however not us. Regardless of the launch, LLMs had no concept we existed on this area.

And that wasn’t our solely drawback.

LLMs have been citing our weblog content material a whole lot of instances. But site visitors to our weblog was falling. Citations confirmed attain, not positioning — an LLM might cite our content material whereas recommending a competitor in the identical reply.

We have been shedding measurable affect whereas our true aggressive place stayed unclear.

That disconnect between citations and affect compelled us to rethink all the things.

So, we used our personal instruments to construct a scientific method to AI visibility. In a single month, we practically tripled our AI share of voice — the share of solutions that point out us versus opponents — from 13% to 32% for our goal prompts.

That was the summer season of 2025. Since then, the self-discipline has moved quick. And so has our method. 

We have re-scoped what we measure, narrowed the place we focus, and realized which techniques truly drive visibility (and which do not).

What follows is the framework we nonetheless use, what’s modified since, the place our visibility stands in the present day, and the right way to run it your self.

The measurement drawback

The measurement drawback is straightforward: Normal attribution cannot see AI affect as a result of LLMs form choices with out sending a click on or a conversion.

We might see LLMs utilizing our content material — citations confirmed that a lot. What we could not see was whether or not they really helpful us, ignored us, or obtained us flawed. Utilization is not positioning, and positioning is what strikes shopping for choices.

The operational facet was more durable. 

Rank monitoring assumes secure positions you may test as soon as every week. AI solutions do not maintain nonetheless: The platforms are non-deterministic, returning completely different responses to the identical immediate inside a single day. In addition they shift quick. In our examine of 230K prompts throughout ChatGPT, AI Mode, and Perplexity, ChatGPT’s citations of Reddit fell from practically 60% of responses to round 10% in a matter of weeks.

So I made a decision we must always change what we measured. 

We stopped monitoring whether or not LLMs used our content material and began monitoring our aggressive place: whether or not we get talked about in any respect, and the way usually versus opponents once we do. (Whether or not AI represents us precisely turned out to matter simply as a lot — however aggressive place got here first.)

The 2 metrics that truly matter 

The 2 metrics that truly matter are visibility and share of voice. Each work like acquainted Search engine optimization metrics, besides they measure affect as a substitute of clicks.

You will be seen in a single reply and nonetheless maintain low share of voice if opponents present up in practically all of them.

We monitor each throughout ChatGPT, Google AI Mode, and AI Overviews, amongst different platforms, utilizing Enterprise AIO.

However the metrics solely imply one thing measured on the suitable prompts. 

Monitoring “AI instruments” tells you little or no. 

Monitoring “finest AI visibility instruments for enterprise groups” tells you whether or not you present up precisely when somebody is selecting an answer. 

Get talked about for prompts like that, and you have entered the customer’s consideration set in the intervening time that issues — with out paying for an advert.

Our optimization framework

We nonetheless use the identical five-step framework we constructed in the beginning. What’s modified is the main focus inside it. This is the method and the right way to replicate it.

Step 1: Establish your goal prompts

Begin by hand-picking the bottom-funnel prompts your staff and stakeholders truly care about — the buying-intent queries the place a purchase order determination is in play.

We started with 39 prompts, like “finest enterprise AI visibility platform,” as a result of they replicate actual shopping for choices. At present we monitor 726. However the greater shift is the combination, not the amount.

We weighted the set towards buying-intent prompts, the place AI is more than likely to advocate a particular device. We additionally stored a smaller set of informational prompts. These not often title any model, so they will not transfer share of voice — however they present whether or not AI treats Semrush as an authority on the broader subject, not only a vendor to listing.

The precept beneath all of it: The prompts you monitor are a tiny pattern of all the things individuals ask.

So we optimize for the intent they characterize, the best way you’d goal a subject in Search engine optimization quite than a single key phrase. Immediate analysis is how we choose those value monitoring.

To trace your personal set, use the customized immediate function in Enterprise AIO. Beginning with out instruments? Take a look at variations manually throughout ChatGPT, Google AI Mode, and Perplexity.

Step 2: Set up baseline measurement

Arrange monitoring and measure the place you stand earlier than altering something.

After we first measured, we sat at 13% share of voice for AI visibility prompts — confirming what we suspected. LLMs did not know we had AI instruments on this area.

That 13% mirrored our authentic 39-prompt set. Because the work matured, we re-scoped to the bigger 726-prompt set, weighted towards more durable, higher-intent queries. The 2 units aren’t straight comparable, so we reset the baseline to match the brand new scope: roughly 15% to start out.

Enterprise AIO automates this, monitoring your visibility and share of voice throughout AI platforms so you are not logging mentions by hand — which will get unmanageable quick.

Then monitor it day by day. 

Bear in mind AI solutions are non-deterministic. Every day information is what tells you whether or not a shift is actual or simply noise. Learn each quantity as a variety — a share of voice that swings between 20% and 40% over a day is regular, so “30% ± 10%” is the sincere option to report it.

Step 3: Audit and improve your current content material

Audit the content material you already personal and discover the pages the place you may naturally strengthen your presence in AI solutions.

The primary transfer is pure product mentions. 

Discover content material that already discusses the issues your instruments resolve. Work them in the place they genuinely match. We had a submit on the right way to get LLMs to say your model. We up to date it with a bit on how Enterprise AIO‘s Supply Impression Evaluation reveals which sources LLMs truly cite — introducing the device precisely the place a reader would need it. 

(No enterprise plan? The AI Visibility Toolkit does the identical at a smaller scale.)

The second transfer is format. 

We have been remodeling dense articles into cleaner buildings — direct solutions up entrance, clear headings, comparability tables — which helps readers and the AI programs that pull from well-organized pages.

The third is the most important shift in how we work. 

After we discover a content material hole, we deepen our protection of the entire subject behind it throughout our owned pages. That means, we present up nevertheless somebody phrases the query.

The check is at all times the identical: The point out ought to assist the reader. If it would not, go away it out.

Step 4: Increase past your area

Your personal web site is not sufficient. LLMs pull from throughout the net — Reddit, Quora, LinkedIn, Medium, and business publications.

At first, Reddit regarded like the apparent place for us to start out. Nevertheless it’s a separate venture in its personal proper, one which wants actual technique, sources, and possession. Its quotation share additionally swings exhausting, as we noticed in our most-cited domains examine, so we have scaled again our focus there whereas we maintain testing what works.

We’re additionally testing LinkedIn and Medium, the place we will publish straight. LinkedIn issues specifically — it is a rising quotation supply throughout AI platforms.

The piece that issues most is accuracy. Among the pages LLMs cite most will not point out you, or will get you flawed — and AI programs reuse that very same context throughout many solutions, so one flawed declare can cascade.

That is why we have began scaling outreach in-house, constructing direct relationships with the homeowners of extremely cited pages. 

We’re after one factor: accuracy. Truthful criticism is one sign AI weighs towards all the things else; an inaccuracy will get repeated throughout reply after reply till it turns into the story. That is the actual danger.

Step 5: Create recent, citable content material

Create new content material that straight solutions your goal prompts, in codecs AI can pull from simply, like listicles and comparisons. Make it authoritative and data-driven: actual solutions backed by specifics.

Listed here are the writing techniques our content material staff makes use of:

• Mirror the heading in your first sentence. If the heading asks “What’s AI visibility?”, open with “AI visibility is …”
• Reply the query utterly in that first sentence, the place readers and LLMs can discover it quick.
• Again claims with specifics. “Cited in 3 of 10 responses for a goal immediate” tells a reader way over “our visibility improved.”
• Skip analogies, idioms, and metaphors. Write “AI visibility is important for discovery,” not “AI visibility is the north star guiding ships by means of digital fog.”
• Maintain antecedents clear. “Enterprise AIO tracks model mentions throughout AI platforms. The device highlights new citations” reads cleanly; “It highlights new citations” leaves readers guessing.
• Select readability over flourish. AI has to know and extract your level quick.

Outcomes: What labored and what did not

The method labored, and it stored working as we raised the bar.

On the more durable 726-prompt set, we have grown our share of voice from 15% to 25%. The positive factors reached past AI-specific matters, too: throughout roughly 1,000 Search engine optimization-related prompts, share of voice rose from 49% to 55% over six months — an indication that sturdy Search engine optimization and AI visibility transfer collectively.

We will not totally isolate which tactic drove what, since we ran them directly, although increasing past our personal area persistently regarded like a serious lever.

Two issues genuinely shocked us. 

The primary was pace — we noticed motion in days, typically hours, far quicker than Search engine optimization. However pace cuts each methods: Content material decays simply as quick, so a web page shedding visibility cannot sit in a backlog. 

The second is what nonetheless would not work — income attribution. Separating AI’s impression from paid search, e-mail, and all the things else is genuinely exhausting. 

The information is enhancing (Microsoft now studies AI efficiency in Bing Webmaster Instruments, together with how usually Copilot cites your pages), however we’re not there but.

What this implies for Search engine optimization groups

We’re all figuring this out collectively, however some classes are already clear.

A 12 months in the past, I tracked weekly rating studies like everybody else. Now I test AI visibility day by day, and I care as a lot about how AI describes us as about the place we rank.

This is what I would inform one other Head of Search engine optimization beginning out:

• Count on your top-funnel content material to lose site visitors. Folks will not click on when AI solutions them straight, so measure visibility, not simply clicks.
• Your personal area is not sufficient, and accuracy issues as a lot as presence. Present up on the websites AI cites, and ensure they characterize you appropriately — a flawed declare spreads throughout solutions.
• Put together stakeholders for brand new metrics earlier than you want funds. Your CEO nonetheless expects site visitors, however your finest outcomes might not present up in Google Analytics.
• Construct content material processes for pace. When visibility drops, the repair cannot wait in a backlog.
• Do not construct your personal AI visibility monitoring. The API prices, maintenance, and information reliability points make purpose-built instruments the higher funding.

That is an extension of Search engine optimization, not a separate self-discipline. The identical fundamentals — authority, clear construction, related content material — resolve whether or not you present up, whether or not an individual or a machine is doing the wanting.

Google now says so outright. In its steering on AI options, Google states that optimizing for generative AI search options continues to be Search engine optimization, and that foundational Search engine optimization is the idea for visibility in its AI experiences.

So the work is not unique. Observe the brand new metrics, cowl your matters the best way good Search engine optimization at all times has, and keep near the info because it shifts.

We’re nonetheless studying. However this a lot is obvious: groups that begin now might be forward when everybody else is scrambling to catch up. Higher to start out measuring than to attend for good readability.