Meridian Gate: AI content at scale, flat pipeline, and a wrong description in assistants
Meridian Gate is a B2B cybersecurity vendor with €60M ARR selling to mid-market IT and security leaders. Nine months ago the content team of four adopted an AI drafting assistant and scaled from 12 to 40 articles a month. Organic sessions are up 55% year on year, but demo requests have stayed flat at roughly 120 a month and sales complains that inbound leads are less qualified.
Analytics shows 1,400 published URLs. 900 of them receive fewer than 10 sessions a month, while 12% of URLs produce 80% of organic-assisted conversions. Legal has also found two articles containing plausible but fabricated statistics, cited to papers that do not exist. Separately, win-loss interviews suggest buyers now ask ChatGPT and Copilot for shortlists, and one assistant describes Meridian Gate as 'a firewall hardware vendor' - a description two years out of date. The CMO has one quarter and no extra headcount.
1. How would you triage the content portfolio and fix the AI content process with no extra headcount?
Read the worked answer
Name the real finding first: traffic up 55% with demos flat means the team scaled supply, not demand capture. More pages is not the problem to solve.
Triage on a two-axis view - organic-assisted conversions versus sessions.
- The 12% of URLs producing 80% of assisted conversions (roughly 168 pages) are the asset. Protect them, refresh quarterly, add practitioner quotes, comparison tables and internal links from everywhere else.
- Of the 900 low-traffic URLs: check backlinks and assisted conversions before touching anything. Most should be consolidated into stronger parent pages and redirected; a small number deserve a rewrite because they target a query buyers actually use.
- The middle band should be judged on assisted conversions and sales mentions, not last-click.
Reallocate, do not add. Cut publishing from 40 to 15-20 pieces a month and spend the freed capacity on depth: buyer-question pages, technical comparisons, integration documentation, and objection-led content that sales can send after a call. Fewer pages that answer a real evaluation question beat forty that answer search queries.
Process fixes that cost no headcount:
- No statistic without a verifiable primary source; the editor clicks the link before publication.
- Named human fact-checker plus subject-matter sign-off on every page; AI drafts outlines only for claim-heavy topics.
- A claim register of approved statistics with dates and owners.
- Monthly spot-audit of 10% of published pages, with a correction log. For the two fabricated pieces, publish a visible correction rather than quietly editing - quiet edits discovered later cost more trust than the original error.
- Change the content KPI from sessions to demo requests, sales-accepted leads and win rate on content-influenced deals.
2. Buyers increasingly ask AI assistants for vendor shortlists. What should Meridian Gate actually do, and how should it measure progress?
Read the worked answer
Be sceptical of easy answers. You cannot buy placement in assistant answers, and most proposed 'answer engine optimisation' tactics are unproven. Spend accordingly.
What is measurable and worth doing:
- Track assistants as a channel. Segment referral traffic from chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini. Volumes are still small but the trend and landing-page behaviour are informative; assistants often send users deeper into the funnel than organic search.
- Run a prompt panel. Write 30-50 realistic buyer prompts - for example, best SIEM for a 500-person manufacturer with EU data residency - and check them monthly across three or four assistants. Record presence, context, and errors. This turns a vague worry into a tracked accuracy rate.
- Fix the factual foundation. Consistent entity data across the site, LinkedIn, Crunchbase, G2, Capterra and Wikipedia where eligible; a one-sentence company description used verbatim; Organization and Product structured data; clean definitional and comparison pages that are easy to quote.
- Correct the record off-site. Assistants repeat what corroborated third-party sources say, not what your homepage claims. Getting 'SaaS platform' rather than 'firewall hardware vendor' into review sites, analyst notes and partner pages is the lever. Treat it as digital PR, not on-site optimisation.
- Do not spend material budget on llms.txt or similar. They are unratified proposals with no guaranteed effect.
Measurement and expectations. Report a quarterly accuracy score from the prompt panel, share of AI referrals, and assisted conversions where assistant sessions touch a demo request. Set a two-quarter horizon for accuracy improvement and avoid promising pipeline from a channel whose mechanics are still opaque.