TL;DR
- Scrunch AI is an AI visibility platform Sitecore acquired in June 2026 for a reported $225 million.
- We bought the $250 plan and ran fifteen unbranded prompts on HubSpot against four rivals.
- It reported HubSpot mentioned in 84.4 percent of responses and cited in 34.4 percent.
- Our own 45-answer check matched Scrunch's numbers within 1.4 points.
- Over three days, 96 responses became 171 and the citation rate went 34.4 to 34.5 percent.
- The $250 Core plan takes a card and auto-bills unless you cancel.
- It covers four AI engines: ChatGPT, Perplexity, Google AI Overviews and Copilot.
- Claude, Gemini, AI Mode, Grok, Meta AI and the Agent Experience Platform are Enterprise only.
- Strengths: citation-level source tables, query fan-outs, competitor detection, and a crawler view.
Scrunch AI is the most expensive tool we have paid to test, and the only one whose owner changed while we were writing about it. Sitecore announced on 3 June 2026 that it had acquired Scrunch. Bloomberg reported the price at about $225 million. Five weeks later the product relaunched as an Agent Experience Platform. We wanted to know what that looks like from inside a paid account.
So we bought the $250 plan, pointed it at HubSpot, and ran it for three days. Here is the short version. Scrunch is the most complete tool in this category we have used, and its measurements survived being checked against our own. The entry plan is $250 a month, it needs a credit card, and it bills you automatically when the seven days end. Four of the nine AI engines are included. The Agent Experience Platform the relaunch was named after is not.
What Scrunch is now
Scrunch started in 2023 and raised $26 million before the acquisition, including a $15 million Series A led by Decibel in July 2025. It says it works with more than 500 brands and agencies, Akamai and Lenovo among them. On G2 it holds 4.6 out of 5 from 73 reviews, and the seller on that profile now reads Sitecore.
The product does three things. It watches what AI engines say about your brand when people ask unbranded questions. It reads your website the way an AI crawler does and scores what it finds. And on the enterprise tier it rewrites what those crawlers receive. That last piece is the Agent Experience Platform, launched on 8 July 2026 alongside two features called Signals and Explorer. If you are new to the category, our guide to what AI visibility means covers the ground this sits on.
How we tested it
We used HubSpot as the test brand, the same one we ran through AthenaHQ, Peec AI and Otterly.AI, so the numbers stay comparable across the series. Fifteen prompts, all unbranded, United States, English, no persona. Five of those prompts were written by Scrunch when the workspace was created. We wrote the other ten.
Then we did something we have not done in this series before. We sent the same fifteen questions to ChatGPT ourselves, three times each, on the same day, and counted the brands in the answers. Forty five answers of our own to set against Scrunch’s. That comparison is further down.
The workspace arrived already built. HubSpot as the brand, Pipedrive, Salesforce and Zoho CRM as competitors, three personas, five prompts, and a crawl of hubspot.com already queued. We added ActiveCampaign as a fourth competitor, switched the country from Global to United States, and pasted in our ten questions. Nothing else.
The dashboard after one day
Ninety six responses came back in the first evening. HubSpot was named in 84.4% of them, its own domain was cited in 34.4%, and sentiment averaged 96.3 out of 100. Across the five brands we tracked, HubSpot led at 84%, then Salesforce at 45%, Zoho CRM at 43% and Pipedrive at 29%.
Below those tiles, the answers were split three ways. Of the responses where HubSpot appeared, 87.7% put it in the top third of the answer and 12.3% in the middle. None landed at the bottom. Sentiment came back positive on 76 responses, mixed on 4 and negative on 1.
This is the part every tool in the category does. What follows is where Scrunch separates itself.
What $250 a month actually buys
The pricing page sells the Core plan on “125 unique prompts”. Inside the product, the meter counts something else: monthly responses, capped at 5,000. Our fifteen prompts showed as 60 active variants consuming roughly 600 responses a month, and the app spells out the sum underneath: 60 variants times 10 responses each.
Work backwards and the 125 figure resolves. One question, run on four engines, ten times a month, is 40 responses. Multiply by 125 and you get exactly 5,000. So the advertised prompt count assumes one language, one persona, one country and four engines.
Add a second persona and the same 125 questions need 10,000 responses, which the plan does not have. The prompt editor shows this before you commit, in a line that reads “1 language x 1 persona x 1 country x 4 platforms = 4 variants”. Every change previews as a diff: adds 4 variants, replaces 20 variants, no change. It is the clearest budget model we have seen in this category, and it also means the headline number on the pricing page is a best case.
One quirk. The plan sheet allows one country, but the editor let us hold Global and United States at the same time and simply doubled the variant count. Nothing blocked it.
Four engines, and the five you cannot have
Core covers ChatGPT, Perplexity, Google AI Overviews and Microsoft Copilot. Open the platform picker and the other five sit there greyed out with the same label attached to each: not available on this plan. Those five are Claude, Gemini, Google AI Mode, Grok and Meta AI.
Two of those absences matter more than the rest. Google AI Mode is where Google is moving its answers, and Claude is the second-largest engine almost nobody optimises for. At $250 a month you get neither. Otterly.AI sells them as add-ons from a $29 base. Scrunch puts them behind a call with sales.
The dashboard filters are less careful about this than the prompt editor. The platform filter on the home screen lists all nine engines, including the five you cannot collect.
We checked its numbers against our own
A visibility tool is only worth its subscription if the percentages it prints are true. So we filtered the Scrunch dashboard down to ChatGPT alone, 24 responses, and set it against our own 45 ChatGPT answers to the same fifteen questions.
| Measure | Scrunch, 24 answers | Ours, 45 answers | Gap |
|---|---|---|---|
| HubSpot mentioned | 79.0% | 86.7% | +7.7 |
| Salesforce mentioned | 42.0% | 57.8% | +15.8 |
| Zoho mentioned | 38.0% | 55.6% | +17.6 |
| Pipedrive mentioned | 29.0% | 40.0% | +11.0 |
| hubspot.com cited | 79.2% | 77.8% | -1.4 |
The brand’s own citation rate lands within 1.4 points. The mention rates all run a little higher in our sample, and none of those gaps is large enough to mean anything at this sample size. On a two-proportion test the biggest of them reaches z = 1.39, well inside the range you get from ChatGPT’s own run-to-run wobble. We measured that wobble directly in our citation variance study: ask the same question four times and only about 12% of sources show up in all four answers.
Nothing we saw suggests Scrunch is reading the answers wrong. It also means a first-week number from any tool in this category is a soft number, because the engine underneath it is noisy. Scrunch’s own answer to that is volume: every variant runs about ten times a month, and the dashboard averages over all of them.
Its prompt set is harder on you than yours would be
Here is the finding we did not expect. Across our own 45 answers, HubSpot appeared in 96.7% of the responses to the ten questions we wrote and 66.7% of the responses to the five Scrunch wrote.
We wrote CRM questions, because HubSpot is a CRM. Scrunch spread its five across content marketing and social media management, and in those answers the shortlist belongs to Buffer, Later, Hootsuite and Sprout Social. HubSpot is absent. Its own dashboard shows the same shape: thirteen of the fifteen prompts mention HubSpot in 100% of responses, and the two that do not are both its own, at 25%.
When we tested AthenaHQ, its auto-generated prompt set included three branded questions and flattered the brand into first place. Scrunch does the reverse. Its default questions reach into categories where the brand loses, which drags the headline percentage down and gives you a more useful picture. Whichever tool you pick, read the prompts it wrote for you before you read the score.
Citations, and the searches behind them
Ninety six responses produced 636 cited URLs. Of those, 10.2% pointed at hubspot.com, 12.9% at a competitor and 76.9% at somebody else. That last figure is the number worth staring at. Three quarters of the sources in answers about CRM software belong to neither HubSpot nor the four brands it competes with.
One thing to watch. The home screen splits all 636 citations by owner and calls the tile Citations. The Citations dashboard counts 510 response-domain pairs instead and reports 6.5%, 9.4% and 84.1% for the same day. That tile does name its unit in the title. The home tile does not, so the two screens will look like they disagree until you notice why.
Each domain carries five numbers: how often it gets cited, how many separate responses it appeared in, how often your brand is mentioned when it is cited, how often a competitor is, and an influence score. hubspot.com scored 34.4%, 33 responses, 100% HubSpot mentions, 36.8% competitor mentions, influence 5.156. zapier.com and thecmo.com sat in the same list, both cited more often than pipedrive.com.
Open a single answer and Scrunch keeps the whole thing: the comparison table, the paragraphs, the source chips under each claim. Two panels sit beside it. One lists every citation in that answer, colour coded by owner. The other is called Query Fan-outs, and it shows the searches the model actually ran.
For the question about tracking customer relationships in one place, ChatGPT ran two searches. The first was “best CRM platforms customer relationship engagement Salesforce HubSpot Zoho Pipedrive Microsoft Dynamics 365 2026”. The second was “CRM customer engagement platform HubSpot Salesforce Zoho Pipedrive official features”. Those are the strings your content has to answer, and they look nothing like the question a person typed. On Core this panel is labelled limited in the plan sheet, so Enterprise sees more of it.
That answer put HubSpot first in its table, “best overall for most small/midsize teams”, with one watch-out attached: costs rise as you add hubs. The eight sources behind it included ciopages.com, meetingnotes.com, technologyadvice.com and a list on tools.davidshadrake.com, which is the sort of long tail our citation research keeps turning up.
It finds the entities you forgot to add
By the end of the first evening Scrunch had eight entity suggestions waiting, each with the evidence attached. Brevo and Microsoft Dynamics 365 as competitors, both seen in four responses at 100% confidence. Then six HubSpot sub-brands: Marketing Hub, Sales Hub, Service Hub, CMS Hub, HubSpot CRM and Operations Hub, each from a single response at 95% to 97%.
Neither Brevo nor Dynamics was in the competitor set we configured. Both are real rivals. This is the feature we would miss most if we moved to another tool, because a competitor set you wrote from memory is always incomplete, and the answers themselves are the only reliable place to learn who you are actually being compared against.
The site audit, and where it stops
We asked Scrunch to audit hubspot.com/products/crm, the HubSpot page cited most often in our own data. It came back in about two minutes with three scores. Access controls 100, two checks passed. Content delivery 100, three checks passed. Content quality 70, two of three checks passed.
Underneath sits the best part: a panel headed “What AI Bots See”, showing the page converted to the plain markdown a crawler receives. Title, meta description, headings, body, nav and footer. Anyone who has argued about whether a JavaScript-heavy page renders for AI crawlers can settle it here in one look. Our guide to optimising pages for LLMs spends several thousand words on questions this screen answers directly.
Two limits. The audit told us content quality failed one check out of three and did not say which one. That detail lives on a deeper tab that only exists for pages already in your site map. And the site map crawl, which started by itself at 21:00, still read “preparing crawl” two hours later. Core caps that crawl at 25 pages per site, which is a reasonable budget for a small site and a rounding error on hubspot.com.
The features with strings attached
Six things on the feature list need more than a login before they show you anything. Worth knowing which, because two of them are gated in ways the pricing page does not spell out.
- Signals flags threshold breaches, trend shifts and emerging competitors, each with a suggested next move. It reads against your own history, so a new workspace shows nothing.
- Prompt Optimization finds prompts inside a topic that return overlapping results, so you can retire the duplicates and free up response budget. It needs fourteen days of data. The trial runs seven, so the card gets charged before you can judge it.
- Agent Traffic reads AI crawler hits from your server logs. It connects through Akamai, Cloudflare, CloudFront, Vercel, a REST feed or a WordPress plugin, and hands you a webhook URL and an API key. Somebody with access to the infrastructure has to wire it in.
- AI Referrals reads referral traffic from AI platforms out of Google Analytics, so it needs the GA4 property for the domain you are tracking.
- Shopping fills only when AI platforms return product results for your prompts. Software brands will find it empty.
- AXP, the Agent Experience Platform the relaunch was named for, carries a dash in the Core column. It is Enterprise only.
Trends sits slightly apart from all of it. It is a separate index of industry topics with their own AI search activity, independent of your prompts. Searching CRM returned two topics: Real Estate CRM Software and Sales pipeline management. Opening the first one, the sentiment panel read no data, the top brands panel read zero of zero, and the competitive presence chart was empty. Below all of that sat three real ChatGPT answers from 15 August, in full, with their citations. The raw material is genuinely interesting. The analysis built on top of it is not finished.
What changed by day three
We bought the plan on 17 August, left the workspace collecting, and came back on the 19th. The sample had nearly doubled, from 96 responses to 171. Almost nothing else moved.
Mentions went from 84.4% to 87.1%. Sentiment went from 96.3 to 96.98. The citation rate went from 34.4% to 34.5%, a tenth of a point across nearly twice the data. The order of the competitors held, with Salesforce and Zoho CRM both on 46.2% and Pipedrive on 32.7%. HubSpot still never landed in the bottom third of an answer, and the count of negative responses stayed at exactly one.
Further up we called a first-week number from any tool in this category a soft number, because the engine underneath it is noisy. On this brand it held. Day one and day three told the same story.
One figure did move. On day one, 636 cited URLs split 10.2% to hubspot.com, 12.9% to a competitor and 76.9% to everyone else. By 19 August that was 1,121 URLs splitting 9.5%, 9.9% and 80.6%. The share belonging to nobody in the competitive set grew as the sample grew.
Signals still had not filled. Widening its window to three months returned zero, so after three days of collection it had produced nothing at all.
The interface changed underneath us as well. A banner now reads “You’re on the new Scrunch experience. Classic is going away soon.” The numbers and the navigation carry across, so nothing above stops being true. It does mean the Classic screens above show an interface Scrunch is in the middle of retiring.
Pricing
| Core | Enterprise | |
|---|---|---|
| Price | $250 a month | Custom |
| AI engines | 4 | 9 |
| Unique prompts | 125 | Custom |
| Countries | 1 | Unlimited |
| Personas | 3 | Custom |
| Competitors | 5 | Custom |
| Site map pages | 25 | Custom |
| Site audits | 5 a month | Full site |
| Page optimisations | 1 a month | Custom |
| AXP | No | Yes |
| API, MCP, CLI, Looker Studio | No | Yes |
| Support | Email, self-serve onboarding | Email and Slack, account team |
The trial is seven days and covers everything in Core. It also requires a card. Scrunch says so in its own FAQ: users are automatically upgraded to a paid subscription when the trial ends unless they cancel. Put a reminder in your calendar for day six.
For context on where that sits, Otterly.AI starts at $29 a month for the same four engines, and SE Ranking folds AI tracking into a full SEO suite. We compared the field in our Profound alternatives test and our best AI visibility tools roundup.
Where it wins, where it does not
It wins on depth. The citation tables, the query fan-outs, the entity detection with evidence attached, the crawler view of your page, the response budget shown as a sum before you spend it. Nothing else we have logged into does all of that in one place. The build quality shows in small things: a destructive bulk edit asks twice, a change that would empty an axis is blocked with the offending rows listed, and removing a variant tells you the history is kept.
It loses on what $250 excludes. Four engines out of nine, no Claude, no Gemini, no AI Mode. A 25-page crawl. One page optimisation a month. No API, no MCP, no Looker Studio. The Agent Experience Platform behind a sales call. If you came for the relaunch, Core is not where it lives.
The gap between those two paragraphs is the decision. Scrunch is priced as an enterprise tool and its Core plan is a cut-down version of one. A small team tracking one brand in one country will get more per dollar from a $29 tracker. A team that needs to know which third-party pages are feeding the answers, and what the model actually searched for, will not find that elsewhere at any price we have seen.
The bottom line
Scrunch measures accurately. We checked, and its numbers held. It also has the most useful screens in this category, and the most confusing gap between what its marketing announces and what its entry plan contains.
Buy it if AI search is a budget line rather than a curiosity, if you need engine coverage beyond the big four, and if somebody on your side can wire up server logs. Skip it if you want a monthly number for one brand, because Otterly.AI gives you that for a tenth of the price. And whichever way you go, read the prompts the tool wrote for you before you trust the score it prints. That is the lesson that keeps repeating across every tool in this category.