NexlaGoogle Ads keyword expansion and 5-page landing page test
Today the account puts most of its MCP money on one generic keyword and sends it to a page that Google rates poorly for relevance. This plan rebuilds MCP into four tight themes, opens two new keyword buckets with a small capped budget, and tests four new landing pages against your current page. Every change waits for your approval.
Momentum 360. Google Ads account 791-780-2207. Monthly cap $2,000. Primary conversion: HubSpot demo request. Nothing in the account has been changed. Numbers are from the Oct 5 account data unless marked as an estimate.
1.9MCP Quality Score, average of 9 scored keywords (Brand is 8.5)
65%of MCP spend sits on one keyword, mcp server platform
10 to 11%MCP impression share, with rank loss bigger than budget loss in 3 of 4 weeks
$1,845planned monthly ceiling at full build, under the $2,000 cap
The gap
We are paying for broad curiosity about MCP and missing the buyers who search for what Nexla actually sells. Four things explain most of it.
One keyword runs the campaign.mcp server platform took $779.75 of $1,195.73 MCP spend at Quality Score 2. The bare term mcp server gets a 0.40% click-through rate, because people searching it mostly want to build or install a server.
Rank is the problem more than budget. More money alone will not fix a Quality Score of 1.9. Relevance has to come first.
Leaks and blind spots. $214.72 of $843.86 visible search-term spend went to developer tools and directories (for example claude mcp, mcp market, playwright mcp). There are no keywords for change data capture, streaming ETL or document ETL that earn impressions today.
We cannot see the result yet. The demo form is not counted as a conversion, and GA4 saw 24 paid sessions against 92 clicks in the week of Sep 28. Both are fixed before any test starts.
Where MCP impressions were lost, by week
Share of eligible impressions lost to rank or to budget
Sep 7
48%44%
Sep 14
64%30%
Sep 21
75%19%
Sep 28
39%50%
Lost to rankLost to budget
Google Ads data, weeks of Sep 7 to Sep 28. Reading the current ad text and the Quality Score components (expected click-through rate, ad relevance, landing page experience) is the first task this week.
New keyword buckets, with volume and cost
We scored 2,543 candidate keywords from your site, search results, related-term research and Keyword Planner. Volume and cost are real for only 248 of them, so the table shows measured numbers only. Unknown does not mean zero.
Keyword Planner, United States. Volume is monthly searches summed over keywords with measured volume. Cost is the median of the low end of the top-of-page range.
Bucket
Candidates
Measured volume per month
Median planner cost per click
When
Why
MCP servers and gateways
563
15,080
$8.61
Now (rebuild)
Core product. Four tight ad groups, exact and phrase only.
ETL, streaming and CDC
324
17,860
$6.47
Oct 27 probe
Zero coverage today. Real demand, so we test it small.
Enterprise data integration
247
46,070
$9.14
Oct 27 probe, narrow
Largest measured demand, but also learner and legacy-tool noise.
System integrations (ERP, CRM, warehouse)
90
10,310
$6.67
Days 31 to 60
Per-system MCP terms. Add selectively after search-term evidence.
Document ETL and RAG
289
2,540
$4.26
Days 31 to 60
Thin volume. Only enterprise variants.
Agent data access and context
230
210
$2.91
Days 31 to 60
Nexla's real thesis, but almost no measured volume. Cheap experiment.
Governance and security
232
0
n/a
Days 31 to 60
Volume unknown. Competitors lead with trust, so we test it.
Competitor alternatives
404
740
$18.78
Days 61 to 90, only with Nexla's written OK
Competitor names as keywords only, never in ad copy.
iPaaS and API integration
83
340
$21.26
Defer
Median planner CPC above $20 on a handful of terms.
Estimate Planner cost is not what you will pay. On the 5 keywords where we have both numbers, your actual cost per click ran 21% to 92% of the planner figure (median 50%), on samples of 2 to 159 clicks. We plan on 0.5x to 0.9x. Planner high ends such as $526 for mcp gateway are ignored. Competitor names stay in keywords and negatives only, never in ad text.
Campaign architecture
Exact and phrase match only. Four to ten keywords per ad group so each ad can say exactly what the search asked for. Dashed cards start later and only after the earlier ones prove relevant.
Google Ads 791-780-2207Search only, United States, weekdays 7 AM to 7 PM CT
Brand
Keep. Cut to $15/day
Brand-Exact (add 6 exact terms)
MCP core, rebuilt
$40/day, build now
M1 MCP Gateway 9 keywords
M2 Enterprise and Governed MCP 10 keywords
M3 MCP Studio and Task-Specific 7 keywords
M4 Agent Access to Enterprise Systems 9 keywords
Data integration and CDC probe
$25/day, launch Oct 27
D1 Change Data Capture 10 keywords
D2 Streaming ETL 7 keywords
D3 Enterprise Data Integration 10 keywords
Agent data long tail
$8/day, Days 31 to 60
L1 Agent Data Layer and Context 7 keywords
L2 Document ETL and RAG 7 keywords
L3 Governed Agent Data 6 keywords
Competitor alternatives
$7/day, Days 61 to 90, only with written OK
Terms proposed when approved
Negative keywords
Twelve themed shared lists, attached by campaign: developer tools, MCP directories, tutorials, free and open source, typos, jobs, academic, legacy ETL products, database admin CDC, competitor brands, competitor support and adjacent products. A first batch is ready for your sign-off. Two traps are already handled: the existing 94-term Generic list may block the new ETL and CDC terms, and a bare negative on Claude would block our own keyword.
Bidding
Manual cost per click per ad group until the demo form is counted cleanly. Smart Bidding needs about 30 conversions in 30 days, and at this volume we expect about 5 validated leads a month, so we say plainly that manual bidding runs through Day 90.
Ad copy samples
Fourteen responsive search ads across seven launch ad groups, two per group. In each group one ad pins a keyword headline in position 1 and leads with proof or scale. The other is unpinned and leads with governance or speed. We change one variable per test cycle.
MCP Gateway, proof angle
Headline 1 pinned
Ad nexla.com/MCP/Gateway
MCP Gateway For Enterprise | Governed MCP Gateway | One MCP Layer, 1,000+ Systems
Task-specific MCP servers cut tokens 32.7x and tool calls 4.7x in Nexla's own tests.
Agentic data integration with streaming ETL and CDC. Over 1 trillion records a month.
Capture changes and stream them to your warehouse or agents in real time.
All 15 headlines and 4 descriptions
Change Data Capture Platform
Enterprise CDC Replication
Real-Time CDC Replication
CDC For Enterprise Databases
Streaming ETL And CDC
High-Volume Streaming ETL
Built For Trillions Of Records
Pipelines That Stay Reliable
Real-Time Data For AI
Book A Nexla Demo
1,000+ Enterprise Connectors
SOC 2 Type II Compliant
1T+ Records Processed Monthly
New Connector In Under 1 Week
Request A Nexla Demo
Agentic data integration with streaming ETL and CDC. Over 1 trillion records a month.
Capture changes and stream them to your warehouse or agents in real time.
Helix learns schemas and metadata to keep pipelines reliable as sources change.
1,000+ connectors including databases and apps. Book a Nexla demo.
All headlines are 30 characters or fewer and descriptions 90 or fewer, checked by script, with no duplicates and no competitor or customer names. Benchmark claims are written as Nexla internal benchmarks and need your legal sign-off before they run in paid ads. Final URLs for data integration ads are left for you to name.
The 5-page landing page test
Four new concept pages and your current page compete for the same MCP traffic. Each page has the same single demo form so the form is never the variable. See the concept pages.
A
Benchmark Lab
Leads with Nexla's four sourced benchmarks, shown as animated proof.
Wins if proof-first copy lifts form starts for evaluation traffic.
B
Context Engine
Explains the right-context-not-more-context idea and the Helix engine.
Wins if explanation lifts engagement for agent and data-access searchers.
C
Enterprise Proof
Security, compliance and customer proof come first.
Wins if governance and trust move enterprise-titled visitors.
D
MCP Studio
A build-it-by-conversation demo as the page spine.
Wins if the closest message match to builder searches lifts CTA clicks.
Control
Current page
nexla.com/lp/mcp-servers/ stays in every round until a challenger clearly beats it.
The baseline everything is judged against.
How traffic is split
A server-side split on a Nexla-controlled address sends each visitor to one page at random, keeps them on it for 30 days, and tags every lead with the page they saw. Google Ads Experiments only handles two arms at once, and splitting by ad is biased, so both are fallbacks.
Before any paid click is routed
GA4 must see at least 80% of ad clicks for 7 days in a row. All five pages must fire the same events and deliver real HubSpot test submissions. Every page passes a mobile and accessibility check.
What this test can and cannot tell you
The MCP test pool is about 35 clicks per week estimate. Detecting a doubling of a 1% demo rate needs 2,319 visitors per page. We will reach about 100 per finalist by Jan 3. So demo submits cannot pick a winner here, and we will not tell you one page won on demos. What the test can do is screen out weak or broken pages and show which message earns engagement and form starts, which we pair with your sales team's view of lead quality.
Phase 1, Oct 27 to Nov 23Five pages at 20% each. Cut a page only for a broken guardrail or no chance on engagement with zero form starts. Keep the current page plus the two best challengers.
Phase 2, Nov 24 to Dec 14Three pages at 33%. Thanksgiving week is read separately. Keep two.
Phase 3, Dec 15 to Jan 3Two pages at 50%. Ship a challenger only if it is at least 90% likely to beat the current page on engagement, does not hurt demo submits, and your sales team sees no drop in lead quality. Otherwise nothing changes. Extension to Feb 7 is your call.
Budget
The ceiling is $85 per weekday, about $1,845 a month at 21.7 weekdays, which leaves roughly $155 under your $2,000 cap for Google's daily over-delivery. Brand drops from $25 to $15 a day because it never lost impressions to budget.
Daily budget per weekday by phase. Bars show the split of the $85 ceiling.
Phase
Split
Per weekday
Per month estimate
Oct 6 to 26
$55
$1,194
Oct 27 to Nov 4
$80
$1,736
Nov 5 to Dec 4
$83
$1,801
Dec 5 to Jan 3
$85
$1,845
BrandMCP coreData integration and CDCLong tailCompetitor
At the Oct 27 budget, MCP gets about 27 to 36 clicks a week estimate at $5.50 to $7.50 each, fewer and better than the 55 a week of generic traffic so far. Any increase above this table is a separate yes from you. We check pacing every Thursday for the first month and raise a flag if month-to-date spend passes 80% of $2,000 before day 22.
What we need Nexla to approve
Nothing goes live without these. We apply only approved items and read each one back from the account with the time it took effect.
This week (Oct 6 to 12)
Approve this plan and the keyword and ad structure.
Count the HubSpot demo request as the single primary conversion. Move the two YouTube actions to secondary.
Approve negative batch 1: developer-tool, directory and typo terms that cost $214.72 of the $843.86 of visible search-term spend since relaunch.
Pause the 40 broad-match MCP keywords and the keyword mcp server platform, which takes 65% of MCP spend at Quality Score 2.
Name an owner for the routing layer (DNS or path) and for the HubSpot hidden field that records the page variant.
Tell us why Google Ads sees 92 clicks but GA4 sees 24 paid sessions in the same week, or give us GA4 and GTM access to find out.
By Oct 19
Sign off on using the benchmark numbers (32.7x, 4.7x, 100%, 25x) in paid ads as "Nexla internal benchmarks", with legal review of the claim wording.
Confirm whether the Gartner Peer Insights and G2 ratings can be quoted.
Approve enhanced conversions for leads.
Name the best existing data integration and CDC landing pages. We did not guess them.
Send the security page and Express.dev page URLs for sitelinks, and confirm whether Express.dev is free to start.
By Oct 26
Approve go-live of the rebuilt MCP campaign.
Confirm the routing layer and HubSpot variant field are live and tested with a real submission.
By Oct 27
Approve the data integration and CDC probe and the move to $80 per weekday. Fallback date is Nov 5.
Later
Nov 23: approve the three-page set after the Phase 1 readout.
Dec 14: written position on competitor bidding, and approve the two finalists.
Jan 3: decide whether to extend the head-to-head to Feb 7.
How we measure: a performing term is one with a validated lead, meaning a HubSpot demo request that passes business-email, bot and company checks and is confirmed by your sales team within 5 business days. Cost per validated lead is reported only when there are at least 3. Until then we report engagement, form starts and search-term relevance, and say "not yet reportable" instead of showing a number we cannot support.