What Does Scrunch AI Actually Sell You?

Most Scrunch AI reviews on page one were written by someone with a horse in the race. Profound published one. Rankability published one and closes by suggesting Rankability. One is an exception worth naming up front: Organik PI bought the $250 plan in August and ran 15 unbranded prompts against it, then checked the tool's numbers against 45 ChatGPT responses by hand. That is real work and it is rare in this category.

We sell no platform. We run the measurement and the work behind it for clients, so the only question we care about is whether the thing you are about to buy can carry the decision you are buying it for.

For Scrunch, that question changed in June, and none of the reviews have caught up with it.

What does Scrunch AI actually sell you now?

Scrunch AI sells two things under one name. The tracker runs your prompts against seven AI platforms and sits on every self-serve tier. AXP, the agent-delivery layer Sitecore paid a reported $225 million for, serves AI crawlers a stripped version of your pages at the edge, and it is Enterprise only.

That last sentence is the whole post.

You can buy the half Scrunch is moving on from. The half it is betting the company on comes with a sales call.

The acquisition changed what the product is for

Every published review still frames Scrunch as a tracker with some extra features. That framing was correct in April. It stopped being correct on June 3.

Sitecore bought the delivery layer, not the dashboard

Sitecore announced the acquisition on June 3, at a figure Bloomberg and Adweek both reported as about $225 million. The press release pairs Scrunch's Agent Experience Platform with Sitecore's own DXP, and CEO Chris Andrew stayed on to run the division.

A content management company did not pay nine figures for prompt tracking. It paid for the part that changes what gets served.

AXP serves agents a different response than it serves your buyers

Scrunch's own launch post, published July 8, describes the mechanism plainly. Universal Optimization "transforms your agent-facing website into token-light, JavaScript-free HTML" by stripping analytics scripts and tracking pixels. Adaptive Optimization goes further and applies the platform's own diagnostic fixes automatically.

The delivery happens, in Scrunch's words, "at the edge." The company is equally clear that the human experience of the site is left completely unchanged.

Read those two claims together and you have the actual product: two versions of your site, one for people and one for machines, kept in sync by a vendor.

The pricing page still sells the tracker, and the tracker is good

Scrunch's published pricing lists Starter at $250 a month billed annually or $300 month to month, with 350 custom prompts, 1,000 industry prompts, 3 personas, 3 seats and 5 page audits. Growth is $417 annually or $500 month to month, with 700 custom prompts, 2,500 industry prompts, 5 personas, 5 seats and 10 page audits.

Every tier names the same seven platforms: ChatGPT, Claude, Gemini, Perplexity, Google AI Mode, AI Overviews and Meta. Seven surfaces on an entry tier is genuinely more than most of this bracket includes.

AXP appears on none of them. Organik PI's hands-on review records the same thing from the inside: a dash in the Core column, Enterprise only.

What a tool comparison asks:

• •How many platforms are included?
• •How many prompts do I get per tier?
• •What does it cost against Profound or Peec?
• •How long does onboarding take?

What an agent-delivery decision asks:

• •Which version of my site does the model actually read?
• •What happens when the two versions disagree?
• •Who owns the divergence when the vendor contract ends?
• •Is the page the reason I am not cited, or is the source pool?

Every published review answers the first list. The second list is the one that decides whether AXP belongs in your stack.

6 questions to settle before you turn an agent-delivery layer on

None of these are Scrunch defects. Each one follows from solving AI visibility at the delivery layer, which is a real lever aimed at one part of the problem.

Question #1: Is your bottleneck token weight, or the source pool?

AXP makes the pages you own cheaper for a model to read. That helps only if the model was going to read your pages.

Our concluded CITE Index study ran 500 buyer prompts nightly through ChatGPT, Gemini and Google AI Mode for 63 days, producing 90,132 answers across 10 consumer categories. Four of the twelve most-cited domains were brand-owned. The other eight were publishers, communities and marketplaces, and no edge configuration reaches a page you do not serve.

Reddit alone drew 14,698 citations and appeared in 13.6% of all answers. That is a larger share than any brand-owned domain managed, and it sits entirely outside your CDN.

Question #2: What happens the first time the two versions drift?

Google has already written the warning for this pattern. Its documentation on dynamic rendering carries a red notice calling the technique "a workaround and not a long-term solution," and recommends server-side rendering, static rendering or hydration instead.

The reasoning is not about ethics. It is about drift. Two rendering paths mean two things to keep current, and the one nobody looks at is the one that goes stale.

We wrote the non-vendor version of this problem in how to run an HTML parity audit. The failure mode there was a page whose answer only existed after hydration. AXP inverts it: the machine-readable version becomes the good one, and your real site is left carrying the debt.

Question #3: Does the cloaking line move because the crawler is an AI agent?

On Google's published definition, no, and the reason matters. Google's spam policies define cloaking as "presenting different content to users and search engines with the intent to manipulate search rankings and mislead users."

Intent and misdirection are both load-bearing in that sentence. Serving the same facts in lighter markup is not the travel-page-for-pharma-page swap the policy describes, and Scrunch is not doing anything of the kind.

The exposure is not today's policy. It is that you have taken on a divergence you now have to police, on a surface where the rules are three years younger than the ones Google wrote.

Question #4: Which crawler are you optimizing for, and does it carry the citation?

The bot fetching your page is often not the system answering the question. Live retrieval, training crawls and user-triggered fetches are separate pipelines with separate user agents, and a rule that fires on one does nothing for the others.

Before you tune anything at the edge, know which agents actually hit you and which of them show up later as citations. We worked through that mapping in which AI crawlers get you cited.

Question #5: What does the page-audit cap actually cover?

Published figures disagree, and the gap is wide enough to matter. Scrunch's pricing page lists 5 page audits on Starter and 10 on Growth. Organik PI's August test reports the Core crawl capped at 25 pages per site, and notes the crawl still read "preparing crawl" two hours in.

Both cannot describe the same limit. Ask on the call whether the number is pages audited, pages crawled, or audits run, and get it in writing before it enters a business case. Any review quoting one figure without the other is untested.

Question #6: Would server-side rendering do the same job with nobody in the path?

This is the question that decides the purchase, and it costs an afternoon to answer. Take the five pages that should be winning your highest-intent prompts, run curl against them, and read what comes back.

If the answer block, the proof and the schema are already in that response, your pages are not token-heavy enough to need a delivery layer. If they are missing, you have a rendering problem that SSR fixes permanently, in your own stack, without a second version of the site to maintain.

A delivery layer changes how your page is served. It does not change whether your page was ever in the running.

What Scrunch does better than the trackers it competes with

Fit is more useful to you than a verdict, and four things here are worth saying plainly.

Seven platforms on the entry tier is the most generous coverage in the self-serve bracket. Peec includes three of six on every tier. Meta is on Scrunch's list and on almost nobody else's.

The industry prompt library is a real head start. A thousand pre-built prompts on Starter means the tool has an opinion about your category on day one, instead of waiting for you to write 350 of your own.

Personas are modelled as a first-class object rather than a filter. Three on Starter and five on Growth is enough to separate the practitioner from the economic buyer, which is where most B2B tracking quietly goes wrong.

And the numbers held up under independent testing. Organik PI compared Scrunch's measurements against 45 hand-checked ChatGPT responses and found they matched within 1.4 points. Very few tools in this category have been checked that way at all, and fewer have passed.

Read what an independent tester found before you read what a competitor published.

Sizing a Scrunch program before you sign

The diagnostic half is done. Here is the sequence we run with clients before they commit to any platform in this bracket, Scrunch included.

Step 1: Read your own source HTML before you buy anything that rewrites it

Run curl against the five pages that should be winning your highest-intent prompts. Check whether the answer block, the proof and the schema are present in the raw response.

If they are, you do not have the problem AXP solves. If they are not, price an SSR fix against an Enterprise contract before assuming the contract is faster.

Step 2: Run your ten highest-intent prompts by hand across all seven platforms

Open ChatGPT, Claude, Gemini, Perplexity, Google AI Mode, AI Overviews and Meta AI, and run the same ten shortlist-stage prompts your buyers ask. Record where you appear, which competitors appear instead, and every domain cited.

Two hours of this tells you which surfaces carry your category. No dashboard replaces that first pass, because it samples your buyers rather than a vendor's panel.

Step 3: Sort the cited domains into publishable and earned

Take the domain list from step 2 and split it in two: sites you can publish to, and sites you have to earn your way into.

If the earned side dominates, and our corpus says it usually does, a delivery layer is the wrong first purchase at any price. Buy against the bigger half.

Step 4: Price the fourth persona and the sixth seat before you compare tiers

Starter gives you 3 personas and 3 seats, Growth gives you 5 and 5. Count the buyer types and the people who will actually open the tool, then check which tier that lands you in before you compare Scrunch's headline price against anyone else's.

Most published comparisons put one vendor's entry price against another vendor's real configuration, which flatters exactly one of them.

Step 5: Establish your noise band before you set any threshold

Freeze your prompt set and run it for four to six weeks with no content or off-page changes, then record the spread. That spread is your category's noise band.

Across our 63 days the category leader flipped on only 18.7% of day pairs, and in four of the ten categories it never changed once. The average category leader held 78.2% of its category's answers. The full corpus is in the final report. Movement that size is not news, and without a band you will hold a weekly meeting about it anyway.

When Scrunch AI is the right buy, and when it is not

Situation	Verdict	Why
You need broad platform coverage on a self-serve budget	Strong fit	Seven platforms including Meta on a $250 entry tier is the widest coverage in this bracket, and it does not move as you climb tiers.
You sell to two or more distinct buyer types	Strong fit	Personas are a first-class object with 3 on Starter and 5 on Growth, which most competitors handle as a filter or not at all.
You are already a Sitecore customer	Watch this closely	The DXP tie-in is the reason the acquisition happened, and the integration path will be better here than for anyone outside that stack.
Your cited-source pool is mostly communities and publishers	Wrong first purchase	Neither half reaches those domains. Fund earned placement first and come back when owned pages are the binding constraint.
You want AXP specifically	Price Enterprise, not Starter	It is on no self-serve tier. The plan you can buy today is the tracker, and the quote is a different conversation.
Your pages already render server-side	Skip the delivery layer	Token-light markup solves a problem you fixed in your own stack, without a second version of the site to keep current.

Each tool in this bracket constrains something different. Peec caps engine coverage at three of six on self-serve. Profound sells sampling depth and the question is how many answers produced each score. AirOps meters output and reaches only pages you publish. Scrunch's constraint is the one buyers keep missing: the half being actively developed is not the half on sale. The rest of the field is mapped in our survey of GEO tooling for 2026, and the six jobs to score any of them against sit in the AI visibility platform buyer's guide.

FAQ

What is Scrunch AI?

Scrunch AI is an AI search visibility platform that tracks how a brand appears in answers from ChatGPT, Claude, Gemini, Perplexity, Google AI Mode, AI Overviews and Meta. It reports brand mentions, citations, sentiment and competitor comparison against a prompt set you define, alongside a library of industry prompts. It also builds the Agent Experience Platform, an edge layer that serves AI crawlers a token-light version of your pages. Sitecore acquired the company in June 2026 and CEO Chris Andrew continues to run it as a division.

How much does Scrunch AI cost?

Scrunch publishes two self-serve tiers and a quoted Enterprise tier. Starter is $250 a month billed annually or $300 month to month, with 350 custom prompts, 1,000 industry prompts, 3 personas, 3 seats and 5 page audits. Growth is $417 annually or $500 month to month, with 700 custom prompts, 2,500 industry prompts, 5 personas, 5 seats and 10 page audits. All tiers name the same seven platforms. AXP is not listed on either self-serve tier. Published figures move often, so confirm against a live quote.

Who owns Scrunch AI now?

Sitecore. The acquisition was announced on June 3, 2026, at a figure Bloomberg and Adweek reported as roughly $225 million, and Scrunch now operates as a Sitecore division with co-founder Chris Andrew still leading it. The stated rationale was pairing Scrunch's Agent Experience Platform with Sitecore's content and experience stack. If you are evaluating Scrunch as a standalone tracker, treat the roadmap as pointed at that integration rather than at the dashboard.

What is the Agent Experience Platform?

AXP is Scrunch's edge layer for AI crawlers. Universal Optimization converts your agent-facing site into token-light, JavaScript-free HTML by removing analytics scripts and tracking pixels, and Adaptive Optimization goes further by applying Scrunch's own diagnostic fixes. Scrunch states the human experience of the site is unchanged, which means two versions of the site exist and must stay in agreement. Google's own documentation calls serving bots a separate rendering path a workaround rather than a long-term solution, and recommends server-side rendering instead.

What are the best Scrunch AI competitors?

The names that come up most are Profound, Peec AI, Otterly, AthenaHQ, Evertune, AirOps, Semrush AI Visibility and Ahrefs Brand Radar. None of them ship an edge-delivery layer, so a straight feature comparison flatters whichever tool you started with. Compare on total cost for the platforms and personas you actually need, then on how many answers sit behind each reported score. If what you want is the delivery half, the honest comparison is not another tracker. It is server-side rendering and an off-page program.

The bottom line

Scrunch built a good tracker, priced it fairly, and had it independently verified to within 1.4 points. Then it sold itself to a CMS company on the strength of a different product, said it was going all in on that product, and left it off the price list.

Do step 1 before you take a demo. If curl already returns your answer block, proof and schema, the delivery layer is solving a problem you do not have, and the tracker is what you are actually shopping for. If it does not, fix the rendering in your own stack where nobody can take it away at renewal.

Then go and do the work no vendor at your edge can reach. Nothing in the CDN earns the third-party mention, wins the comparison-page slot, or puts you in the subreddit your buyers actually read. That gap is why a managed GEO agency sits next to tooling rather than inside it, and an AI visibility audit will show you which half your gap sits in before you sign for a year of anything.