The ten improvements
Each entry: what to do, the evidence behind it, and the expected effect. “Ship existing” means the work is already on an unshipped branch or disabled on main.
1Rebuild the corridor page class into real guides, Gulf first
Country to countryEffort L~70% ship existing
Port the geo branch’s flat-file corridor content system onto current main (cherry-pick, not merge: main is 344 commits ahead). Scrub before shipping: 17 lines of exact regulatory thresholds that violate INV-GEO-010, and 99 “compare providers” phrasings. Ship the corridor half of the schema branch with it, so the FAQ accordion is visible on the page and string-matches the FAQPage markup, replacing the broken invisible schema. Write the ~12 missing priority corridors (UK to UAE, UK to Saudi, UAE and Saudi to EU destinations, UK and US to Israel). Retitle to how people actually search and to large-amount framing: “Send Money from the UAE (Dubai) to the UK: Large Transfers” at 61 characters, instead of the current 84-character title that truncates in results. Named author and real dates on every page. Roll out behind the branch’s existing feature flag, Gulf first, with a humanizer pass on all copy.
Evidence
All 1,890 corridor pages share one 202-word body; the text is literally identical between corridors with the country names swapped. CT is absent from all 14 Gulf results sampled. The pages that win (Cambridge, Key Currency) are 2,500-4,500-word authored, dated guides.
Impact: the single biggest lever. From unable-to-compete to top-10 on Gulf head terms and top-5 on long-tail within roughly 3-6 months of recrawl.
2Get into the pages AI assistants actually cite, and give them a reason
GEO, off-siteEffort M, ongoing
Inclusion campaign in order of likely yield. Good Money Guide lists brokers with 13 Trustpilot reviews while CT has ~1,080, and it runs an “Apply To Be Featured” page: apply, enter the 2027 awards, offer a CEO interview. TopMoneyCompare ranks partly on Trustpilot scores, where CT’s 4.8 beats OFX’s listed 4.2. Finder UK: extend the existing Finder Canada partnership and refresh their stale 2023 review. Enter Moneyfacts. Correct the record where AI reads lies: MoverDB’s false FCA claim, Finder Canada’s “28 currencies, no app”. Refresh Wikipedia and Wikidata via disclosed talk-page requests (the infobox leads with Daniel; the newest source is from 2015). Then give everyone a reason to cite CT: a quarterly High-Value FX Transfer Report built from CT’s own anonymised data (median margin by size band and corridor, settlement times, forward usage). Original statistics are what listicles and LLMs cite, and nobody in this niche publishes any. Finally, restart review velocity with post-trade Trustpilot invitations: only 24 reviews arrived in the last 12 months, and review recency is a ranking input on the listicles themselves.
Evidence
Zero of eight listicles list CT; zero of ten AI-retrievable category queries return it; AI answer summaries already quote CT-style positioning from competitors’ pages. No press since around 2016, no industry awards ever entered.
Impact: the biggest pure-GEO lever. Listicle presence from 0 of 8 to 3-4 of 8 within two quarters; AI category mentions from roughly zero to measurable. Set a 20-prompt baseline panel before shipping anything, so the change is provable.
3Make the crawl layer tell the truth
Site-wide, technicalEffort S-MPart ship existing
Delete the static 2018 currency sitemap sitting in the prod docroot: it shadows the dynamic generator, omits AED entirely, and lists retired currencies at non-canonical uppercase URLs. Fix the generators to emit lowercase, no-slash canonical URLs with real per-page lastmod (today the timestamp is generated at request time, so it changes between fetches), and to submit only indexable pages. Ship the schema branch commit that 410s invalid currency slugs and 301s junk paths, and extend the same treatment to /countries/ and to the unanchored rewrite that currently makes nonsense URLs resolve as indexable self-canonical duplicates. Curate the ~9,000-13,000 indexable amount pages down to 3-4 large tiers per priority pair, and noindex every sub-£5k tier. Kill the TransferWise affiliate pages. De-slash the ~88 redirecting internal links per pair page, fix the 90 dead links on /countries, add HSTS.
Evidence
Zero of 2,702 programmatic sitemap URLs resolve as canonical 200 pages today; every corridor URL in the sitemap 301s via a trailing slash. Three agents converged on this independently.
Impact: the enabler that decides whether gains from items 1 and 4 land. Canonical share of submitted URLs from 0% to above 90%, duplicate inventory from unbounded to near zero, crawl budget concentrated on ~1,500 money pages, and trusted lastmod means faster recrawl of improved pages.
4Retarget currency-pair pages to large-amount intent, with fresh rates
Currency to currencyEffort M-LBuild on existing pattern
Rebuild roughly 24 priority pairs (both directions of the critical corridors) on main’s existing tiered-content pattern: AED to GBP is already ~40% unique while most pairs are 2-7% unique template filler. Move titles off converter intent, which Google, XE and Wise own, onto transfer intent: “AED to GBP Transfers: Live Rate and Quotes on AED 50,000+”. Add worked margin-cost examples at 100k, 500k and 1m, document requirements in “varies by bank” language, and forwards and market orders for completion dates. Fix rate freshness: port the branch’s real rate-timestamp to replace the fake “last updated” render date, and drop the Cloudflare edge TTL on currency and country pages to 1-6 hours. Fix the UAE weekend error. Strip the comparison and small-transfer framing (“From small to large transfers”, “Under AED 25,000…”). Do not merge the geo branch’s currency-page copy: its FAQs are generic and off-positioning. Cherry-pick the rate-timestamp work and the links module only.
Evidence
CT’s only ranking URLs anywhere in the 41 sampled results are two large-amount pair pages, proof the intent works where the content exists. Rates shown to crawlers are 3.4 days stale behind the page cache; Key Currency stamps theirs to the minute.
Impact: top-10 on large-amount long-tail across priority pairs, and the first AI citations for rate-flavoured questions, since answer engines quote passages with concrete amounts and named banks.
5Finish the property launch: the cluster is invisible by accident
New pagesEffort S-MMostly ship existing
Ship the property hub with a full launch checklist. Repoint nav, homepage and the 18 property blog posts that still link the legacy page, whose copy is off-positioning. Add the 301 from legacy to hub (not yet coded on the branch). Move titles into query language: none of the three live page titles contains the word “property”, and the hero-copy sign-off left titles and metas as Monday’s open item, so bundle them. Enable schema on LP-mode pages: LP mode currently disables the schema builder, so all three live pages carry zero JSON-LD; add FAQPage from the existing Q&A content plus WebPage and Breadcrumb. Publish the ten finished guides already on main (4,500-6,600 words each, 107 FAQs between them) as country spokes, after an INV-GEO-010 pass and named authors. Cross-link pairs, corridors and property. Add property URLs to llms.txt, purge the cache, and close the staging leak.
Evidence
Zero internal links point at the three live pages from anywhere on the site; even exact-title queries don’t return them. The open gaps by search evidence: Israel for US buyers (no FX firm in 9 results), UK property bought from the Gulf (none in 9), Spain how-to-pay intent underserved. Portugal, Greece and Cyprus are FX-saturated; skip them.
Impact: from three orphaned pages to a linked cluster of 14 or more targeting results with weak or zero FX competition, in the highest value-per-lead segment on the site.
6Build the high-value use-case portfolio where specialists already win
Other pagesEffort MPart ship existing
Four clusters. Gulf exit events: rebuild the Saudi expat guide, today ~450 words of lifestyle tips with no NEOM, no end-of-service gratuity, no peg, into “Leaving Saudi: moving savings and your end-of-service award home”, with UAE (gratuity, DEWS) and Qatar twins. Large-amount pillar: upgrade the large-money-transfers page, currently 1,300 generic words, into the “£100k+ abroad” pillar with £100k, £250k, £500k and £1m tier guides covering source-of-funds, staging and forwards, which nobody in those fragmented results provides. Inheritance corridor guides: a hub plus USA to UK, UK to USA, Australia to UK and Spain to UK; estates settle over 6-18 months, a natural forward-contract fit, with tax strictly deferred to HMRC and advisers. Bank-alternative pages: switch on the page class that has been built and disabled on main since January, reframed as “what £10k-£500k really costs at {bank}”, and add Gulf banks (Emirates NBD, ADCB, FAB, Al Rajhi, SNB).
Evidence
Cambridge’s 4,300-word gratuity guide is the only specialist present in the Saudi exit results, and Cambridge is #1 for UK-USA inheritance queries. Key Currency ranks for “HSBC international transfer fees” with exactly this bank-page type at small-broker authority.
Impact: four to six clusters with high win-probability and very high value per lead; the fastest top-5 wins on the exit-event long-tail.
7One truth for the entity: founder, dates, facts, authors
Site-wide, GEO + E-E-A-TEffort MShip branch + one decision
Ship the About refresh after fixing its own bugs: it says founded 2010 while llms.txt and Wikipedia say 2014, and it still contains placeholder text and a “comparison site” line. Remove Daniel as active everywhere; AI search summaries currently answer that he is co-founder and CEO. Collapse the homepage’s two contradictory Organization nodes and two WebSite nodes (founded 2013 versus 2014, different logos and descriptions) into one graph with Stevan Litobac as sole founder and full sameAs links (Wikipedia, Wikidata, Crunchbase, Companies House, App Store, LinkedIn). Fix author chaos: the visible author, the JSON-LD author and the microdata disagree on 72% of 315 posts; align them and give the real experts indexable author pages with Person schema and reviewed-by dates. Reconcile “zero markups” with the margin disclosure, and the meta “4.9” rating with the actual 4.8. Trust pages: /partners promises regulatory details and shows none, so add regulator and FRN per partner; /security-regulation must state plainly that CT itself is not FCA-authorised; render the Trustpilot score as real text rather than CSS content, which extractors cannot read. And decide mycurrencytransfer.com: the old “comparison site” is live, indexed, outranks CT for branded corridor queries, and is teaching LLMs the wrong entity. Page-level 301s into CT, keeping the exchange-rates API host, or at minimum de-optimise it.
Impact: branded and trust answers go from four or more conflicting fact-sets to one, and a YMYL E-E-A-T floor goes under every content play above.
8Point the blog’s engine at the money
Other pagesEffort MMechanical + editorial
The content engine publishes about 23 posts a month, but 89% of 2026 output (195 of 218 posts) is robots-blocked market commentary, still submitted in sitemaps (~1,186 blocked URLs) while the AI-bot robots groups allow it: robots groups don’t merge, so Googlebot and Bingbot are blocked where GPTBot is allowed, which also hides the content from ChatGPT’s Bing-backed search. Fix the mechanism in order: remove the briefs from sitemaps, then swap the Disallow for noindex,follow. Launch four to six indexable evergreen weekly outlook pages (GBP/EUR, GBP/USD and so on, bylined by the real analyst) targeting forecast queries. Run an internal-links pass adding 3-5 contextual money-page links per post across ~290 posts, with topic-matched CTAs. Prune the wrong-intent posts (traveller currency explainers, digital-nomad listicles, the remittance-framed “cheapest way” post), merge the duplicate Australia guides, retitle and 301 the UK-US blog post that cannibalises its corridor page, and retire the Western Union and WorldRemit remittance-framed pages along with the banned footer label.
Evidence
27 of 27 sampled 2025 posts contain zero links to money pages. A 2,092-word Portugal property post carries 29 external links and none to CT’s own property pages.
Impact: over a thousand new contextual internal links to money pages, blocked-URL submissions to zero, and a content budget that starts compounding instead of evaporating daily.
9Give the site a spine: origin hubs, a real index, cross-class links
Site-wide architectureEffort MBuild + branch reuse
Build origin hubs for the UAE, Saudi Arabia and the UK first; it is the page type competitors run (XE’s send-money-from pages, Cambridge’s Middle East hub) and /send-money-from-uae/ currently 404s. Rebuild /countries, which is linked site-wide, as a real index grouped by origin; today it renders a duplicate of the UK-US page containing 90 malformed dead links. Define the linking rules: corridor to matching pair in-body with no-slash URLs; corridor to property, exit-guide or pension pages where relevant; pair to property by currency (EUR to Spain, ILS to Israel, AED to Dubai); hub to top corridors. Replace the 158-link dropdown mesh, which links noindexed same-currency pairs, with a form, and settle the nav semantics: pair page for rates and amounts, corridor page for the guide. Reuse the geo branch’s internal-links module.
Impact: site-wide link equity flows to priority money pages instead of a duplicate page with dead links, and the hubs give LLMs an entity-level “sending money from the UAE” page to cite.
10Clear the render path
Site-wide, performanceEffort MBuild
VWO’s anti-flicker snippet white-screens every page for up to two seconds, and two A/B stacks (VWO and Convert) load everywhere; scope VWO to pages with live tests or element-level hiding, and drop Convert where unused. The blog hero’s largest-contentful-paint element is a lazy-load placeholder GIF marked fetchpriority=high while the real image waits for jQuery; switch to native lazy loading and exclude the hero. Apply the homepage’s existing defer pattern to the property and blog templates, which carry 5-7 render-blocking stylesheets. Remove the IE8 shims and unused jQuery UI. Tell Cloudflare to ignore utm parameters in the cache key: today a utm parameter means a cache miss, 343ms versus 55ms time-to-first-byte.
Evidence
PageSpeed Insights refused keyless API calls this session, so no field data was pulled; the body-hide and hero bugs are visible in the page source regardless. Pull the Search Console Core Web Vitals report to quantify before and after.
Impact: honest framing: a supporting lever, not a primary one. But the two-second body-hide is the single most likely real-user LCP drag on slow mobile, and Gulf audiences skew mobile.